Machine learning driven optimization of biofuel powered urban mobility for cleaner and smarter cities

1 Introduction

Over the last few decades, the surge in urbanization and the demand for transportation has greatly contributed to air pollution and Greenhouse gases emissions globally. Transport contributes approximately 13–15% of India’s total greenhouse gas emissions, while major cities such as Delhi frequently experience AQI levels categorized as Poor, Very Poor, and Severe. In the rapidly growing urban transportation systems in India, carbon dioxide emissions, energy use and environmental pollution have come into one of the biggest contributors. With the growing number of vehicles, population growth, industrial development and reliance on fossil fuel based transport systems, there has been a rise in concerns about climate change, poor air quality, and public health issues. Delhi, Mumbai and Bengaluru among major cities often have ‘hazardous’ Air Quality Index (AQI) (Rautela and Goyal, 2024; McKinney, 2010) readings as a result of high level of vehicular emissions and traffic jams. Pollutants from vehicles such as Unburned Hydrocarbons, particulate matter (PM₂.₅ and PM₁₀), carbon monoxide, carbon dioxide and Nitrogen Oxides have a significant impact in the development of respiratory diseases, cardiovascular diseases, and environmental instability. Although various governmental measures and environmental policies have been implemented so far to lower emissions resulting from the transportation sector, urban mobility systems have not yet been sufficiently developed to utilize less conventional fossil fuels (Chauhan & Mishra, 2025; Chauhan et al., 2026a; Chauhan et al., 2026b). Various initiatives like the National Biofuel Policy and Faster Adoption and Manufacturing of Electric Vehicles (FAME) Scheme have driven the adoption of cleaner alternatives like biofuels, electric vehicles, and hybrid solutions. Some of these options are particularly interesting, including biofuels like ethanol based petrol and biofuel (biodiesel) because they can be used straight in existing internal combustion engine systems, with limited modification (mahdnejad, 2025). Biofuels offer a potentially promising route to reduce the net Greenhouse gases emissions resulting from combustion as the carbon dioxide produced in combustion is partly compensated by carbon dioxides taken up when growing the biomass. Ethanol blending programmes such as India’s E20 fuel programme will be expected to cut down lifecycle emissions and minimize reliance on imported fossil fuels (Andrade et al., 2024; Arora et al., 2022).

The environmental challenges associated with urban transportation can be broadly categorized into greenhouse gas emissions and local air pollutants. Greenhouse gases such as carbon dioxide (CO₂) contribute to global climate change by increasing atmospheric heat retention, whereas local pollutants including particulate matter (PM₂.₅ and PM₁₀), nitrogen oxides (NO_x), sulphur dioxide (SO₂), carbon monoxide (CO), and ground level ozone directly affect human health and urban air quality. Although these indicators are related to transportation activities, they represent distinct environmental dimensions and therefore require separate monitoring and assessment approaches for sustainable urban mobility planning (Wang et al., 2025). Beyond India, several international policy initiatives have accelerated the transition toward cleaner transportation systems and alternative fuels. The European Green Deal, Renewable Energy Directive II (RED II), Renewable Energy Directive III (RED III), and the Alternative Fuels Infrastructure Regulation (AFIR) have established ambitious targets for greenhouse gas reduction, renewable energy adoption, and sustainable transport infrastructure development. These initiatives emphasize the growing global importance of biofuels, alternative propulsion systems, and intelligent transportation technologies in achieving long term environmental sustainability and carbon neutral mobility.

The environmental advantages of biofuels, however, depend on a number of dynamic factors, such as the properties of the engine, fuel blend ratios, driving style, traffic volume, environmental conditions and maintenance practices. Typically, the emission estimation methods are based on static emission factors, deterministic models or on a predefined emission average in the regulations, which do not reflect the nonlinear and complex relationships between these factors. These methods are not flexible to the urban context and cannot offer accurate and real time predictions to support sustainable transportation planning and emission management. To overcome these challenges, machine learning (ML) has become an important data driven solution that learns the complex relationships directly from real world data (Sun et al., 2021; Ahmed et al., 2023). The modeling ability of ML algorithms like Linear Regression, Decision Trees, Random Forest Regression, Gradient Boosting, and Neural Networks enable the modeling of nonlinear interactions between vehicle characteristics, fuel properties, operational conditions, environmental parameters and emission outputs. These models can predict with high accuracy emissions of CO₂, fuel consumption and pollution, which can facilitate the dynamic optimization of urban transportation systems (Kumar et al., 2024).

Moreover, the incorporation of the Internet of Things (IoT) sensors, GPS systems, real time traffic data and environmental monitoring systems allows for the gathering of data and making decisions at any time, in real time, for intelligent transportation systems. By combining machine learning with biofuel powered transportation systems, new opportunities are opened in sustainable urban mobility and smart city development. To ensure the efficiency and convenience of transportation while still addressing environmental issues, intelligent frameworks can be employed to assist in real time emission forecasting, fuel optimization, route planning, and environmental monitoring. Advanced visualization web based platform and its analytical platforms can then help the policymakers, urban planners and environmental agencies in finding pollution hotspots, analysing trends in AQI and taking informed measures to reduce emissions accordingly (Kim et al., 2022; Sahu et al., 2022). Hence, this study suggests a framework that is based on machine learning for optimizing urban mobility in the major Indian metropolitan cities using biofuels to mitigate CO₂ emissions and enhance urban air quality.

The system proposed includes modeling techniques, the simulation of the environment, and interactive visualization tools to assess the influence of various fuel types, vehicle configurations, and traffic scenarios on CO₂ emissions, AQI levels, and the resulting air quality. It integrates various data sources such as vehicle emission records, environmental metrics, socioeconomic factors, and policy driven fuel standards, enabling a holistic and contextual understanding of sustainable transportation systems. The primary scientific contribution of this study is the development of an integrated decision support framework that combines machine learning based CO₂ emission prediction, biofuel scenario evaluation, air quality monitoring, environmental simulation, and interactive visualization within a unified smart city platform. Unlike existing studies that investigate these components independently, the proposed framework enables simultaneous assessment of transportation emissions, alternative fuels, and urban environmental conditions to support data driven sustainable mobility planning. The proposed model integrates all the above features on a single platform, i.e., predictive emission analytics, biofuel optimization, AQI analysis and smart city visualization. This contribution is further enhanced by the following key advances:

• A machine learning based emission prediction model that can simulate nonlinear relationships between vehicle specification, fuel blends, operational parameters and environmental conditions.
• Predicting CO₂ emissions, analysing AQI, visualizing pollution hotspot and smart city analytics in an integrated environmental monitoring system.
• An online interactive web platform to visualize, support decision making and evaluate sustainable urban mobility in real time, based on multiple data sources and predictive analytics.

The proposed framework integrates emission prediction, air quality modelling and optimization of biofuels into a single scalable, adaptive framework, not just simultaneously considering them individually, as existing studies have done. This work is a fusion of machine learning, environmental simulation and intelligent visualization to aid real time decision making in sustainable transportation planning and smart city management. An overview of the working of the web based platform can be seen in Figure 1 it shows the flow of the proposed ML Driven Web Platform named Biofuel Powered Urban Mobility for Smart city transportation in order to achieve sustainable mobility (Wang et al., 2025). It demonstrates the process of gathering data from various sources related to vehicles, environment, emissions, air quality, and biofuels, cleaning the data, engineering new features, and applying machine learning to analyze and leverage the data. The Random Forest Regressor predicts vehicular CO₂ emissions based on vehicle and fuel characteristics, while the biofuel analysis module evaluates alternative fuel scenarios to estimate potential emission reductions. The results are subsequently integrated into the web based platform for visualization, environmental assessment, and decision support analysis.

1.1 Statement of significance

• Problem or Issue: The transport sector makes a major contribution to CO₂ emissions, air pollution and poor air quality, particularly in fast growing cities of India.
• What is New and Significant: Machine learning has been used in past studies to predict emissions, forecast air quality, and analyze biofuels. Most approaches however, deal with individual aspects and lack a single framework of integrating transportation with environmental monitoring and smart city analytics.
• What this Paper Brings: This research introduces a novel framework that combines all four aspects, CO₂ emission prediction, biofuel optimization, AQI monitoring, and interactive visualization, within a unified platform through the use of machine learning.
• Who would benefit: The platform can further assist policymakers in evaluating the environmental implications of alternative fuel policies, identifying pollution hotspots, prioritizing mitigation strategies, and supporting evidence based urban transportation planning.
• Research gap: Existing studies have separately investigated vehicular emission prediction, air quality forecasting, biofuel performance assessment, and intelligent transportation systems. However, there remains a lack of integrated frameworks capable of simultaneously combining machine learning driven emission prediction, biofuel optimization, real time environmental monitoring, and interactive decision support visualization. This fragmentation limits the practical application of research outcomes in smart city transportation planning and environmental management.

This paper is organized as follows: The literature survey that is discussed in Section 2 covers the research work on machine learning based emission prediction, biofuel optimization, air quality assessment and intelligent transportation systems. In Section 3, the proposed methodology for this work, which involves data collection, preprocessing, feature engineering, machine learning models, and system architecture, is described. The proposed framework is discussed with the results and performance evaluation in Section 4, which consists of prediction of CO₂ emission, analysis of AQI, and the output of visualization. Finally, Section 5 summarizes the study and points out research avenues for sustainable urban mobility and intelligent environmental management. Overall, the platform can help policymakers, researchers, urban planners and students make informed decisions to improve urban air quality and cleaner transportation by using data.

To address the identified research gap, this study seeks to answer the following research questions:

RQ1: Can machine learning models accurately predict vehicular CO₂ emissions under varying vehicle and fuel conditions?

RQ2: How do alternative biofuel options such as E10, E20, B20, and Bio CNG influence estimated transportation related emissions?

RQ3: Can the integration of emission prediction, air quality monitoring, and interactive visualization improve decision support capabilities for sustainable urban mobility planning?

The novelty of the proposed Biofuel Powered Urban Mobility Platform (BUMP) lies in its integration of machine learning based CO₂ emission prediction, real time AQI monitoring, biofuel optimization, environmental simulation, and interactive smart city visualization within a single decision support framework. Unlike existing studies that focus on individual aspects of transportation sustainability, the proposed platform provides a comprehensive approach for environmental monitoring and sustainable mobility planning. The objectives of this study are:

• To develop an integrated Biofuel Powered Urban Mobility Platform (BUMP) for sustainable transportation analysis.
• To predict vehicular CO₂ emissions using machine learning techniques.
• To evaluate the environmental benefits of alternative biofuels including E10, E20, B20, and Bio CNG.
• To monitor and visualize urban air quality conditions using real time AQI data.
• To support policymakers and urban planners through intelligent environmental decision support tools.

2 Literature review

Urban transportation sustainability has emerged as a multidisciplinary research area involving emission prediction, air quality assessment, alternative fuels, intelligent transportation systems, and smart city technologies. Recent advances in machine learning, Internet of Things (IoT) infrastructure, environmental sensing, and transportation analytics have enabled the development of data driven approaches for monitoring and mitigating transportation related environmental impacts. However, despite substantial progress in individual domains, the integration of emission prediction, air quality monitoring, biofuel evaluation, and decision support visualization remains limited. This section critically reviews the existing literature and identifies the research gap addressed by the proposed Biofuel Powered Urban Mobility Platform (BUMP).

2.1 Machine learning based emission prediction

Although machine learning models have demonstrated superior predictive capabilities compared with traditional deterministic approaches, most studies primarily focus on prediction accuracy and model optimization. Limited attention has been given to integrating emission prediction outputs into broader environmental monitoring and transportation decision support systems (MDPI, n.d.). Consequently, the practical application of these models in sustainable urban mobility planning remains constrained.

2.2 Air quality index (AQI) prediction

Despite significant advances in AQI forecasting, most existing frameworks focus exclusively on pollutant prediction and public health risk assessment. The relationship between transportation related emissions, fuel selection strategies, and air quality management is often treated separately (Harris et al., 2020; Li et al., 2021; Milojević et al., 2025b). As a result, current AQI prediction systems provide limited support for integrated transportation sustainability planning.

2.3 Biofuels for sustainable transportation

Recent studies have highlighted the environmental and operational benefits of biodiesel blends, biomethane, and other renewable fuels in transportation systems. Biodiesel blends such as B20 have demonstrated the potential to reduce lifecycle greenhouse gas emissions while maintaining acceptable engine performance and thermal efficiency. Similarly, biomethane and Bio CNG have emerged as promising alternatives for public transportation fleets due to their lower carbon footprint and compatibility with existing natural gas infrastructure (Chen et al., 2022; Khatri and Kumar, 2024; Milojević et al., 2025a). However, the environmental benefits achieved by these fuels depend strongly on operating conditions, vehicle characteristics, maintenance practices, and fuel blend composition. Therefore, intelligent analytical platforms capable of evaluating alternative fuel scenarios under varying transportation conditions are required to support sustainable mobility planning and policy development.

2.4 Smart mobility and intelligent transportation systems

With the advent of Intelligent Transportation Systems (ITS), cities may take advantage of real time data, predictive analytics, and simulation to enhance the efficiency and environmental sustainability of their transportation systems (Wang et al., 2020; Sun et al., 2021). The modern platforms of ITS combine traffic monitoring, environmental sensing and decision support systems to solve urban mobility issues. The studies have also recently started to include the use of visualization tools and web interfaces that give information about pollution hotspots, emission trends, traffic congestion pattern, mitigation measures, etc. These technologies aid in data driven decision making and help develop smart cities by helping with adaptive transportation management in dynamic urban environments. Recent developments in autonomous and connected vehicle technologies have further expanded the scope of sustainable urban mobility. (Milojević et al., 2018; Milojević et al., 2026) Autonomous electric and hydrogen powered transportation systems are increasingly being investigated as future solutions for reducing emissions, improving operational efficiency, and enhancing transportation safety. Although such technologies remain outside the immediate scope of the present study, their integration with intelligent environmental monitoring and predictive transportation platforms represents an important future research direction for smart city development (Madkhali et al., 2026; Minhass et al., 2026; Mohan et al., 2026).

2.4.1 International policies and sustainable transportation

The transition toward sustainable transportation is increasingly supported by international policy frameworks. The European Green Deal has established ambitious targets for carbon neutrality and greenhouse gas reduction across transportation sectors. Furthermore, the Renewable Energy Directive II (RED II) and Renewable Energy Directive III (RED III) promote the adoption of renewable fuels and advanced biofuels in transportation systems. The Alternative Fuels Infrastructure Regulation (AFIR) further supports the deployment of alternative fuel infrastructure to accelerate transportation decarbonization. These initiatives demonstrate the growing importance of integrating alternative fuels, intelligent transportation systems, and environmental monitoring technologies within future smart city transportation frameworks.

Overall, the reviewed literature confirms that machine learning, alternative fuels, air quality monitoring, and intelligent transportation systems each contribute significantly to sustainable mobility. Nevertheless, the lack of integration among these domains limits their practical applicability for comprehensive environmental decision support in smart city transportation systems.

2.5 Research gap

The existing literature demonstrates significant advances in machine learning based emission prediction, air quality forecasting, biofuel assessment, and intelligent transportation systems. However, these research areas have largely evolved independently. Emission prediction studies primarily emphasize algorithm development and predictive accuracy, while AQI forecasting studies focus on pollutant monitoring and public health applications. Biofuel related research generally evaluates fuel performance and emission reduction potential, whereas intelligent transportation studies emphasize traffic management and mobility optimization. As a result, a substantial gap remains in the development of integrated frameworks capable of simultaneously combining machine learning driven emission prediction, biofuel scenario evaluation, air quality monitoring, environmental simulation, and interactive decision support visualization. Existing smart city platforms rarely provide a unified environment that allows transportation stakeholders to assess environmental impacts, compare alternative fuel strategies, and visualize sustainability indicators in real time.

To address this gap, the present study proposes the Biofuel Powered Urban Mobility Platform (BUMP), which integrates CO₂ emission prediction, biofuel optimization, AQI monitoring, environmental simulation, and interactive visualization within a single machine learning driven framework designed to support sustainable urban transportation planning and smart city decision making. Below we can see Table 1 with the information about Comparative Literature Review and Research Gap Analysis.

Table 1. Comparative literature review and research gap analysis.

3 Materials and methods

The research approach adopted in this study aims to develop a machine learning based framework to maximize biofuel based urban mobility and reduce carbon emissions. The method combines various datasets of different scales and types, including vehicle level emission statistics, region specific social economic data, and sector wise urban emission records, to offer a comprehensive and multi scale analysis. The suggested scheme combines several data sources to enhance the accuracy of the emission prediction and environmental monitoring. Machine learning formula coefficients were derived using the Canada CO₂ Emission Dataset from Kaggle, and the Kaggle India State Emissions dataset was used for the emission information for different regions of India. City level emission distribution data was provided by the CHETNA dataset, and real time AQI values were provided by the Open Weather Map API. Further, the India National Biofuel Policy was referred for the percentage reduction in emissions for E20/B20 and the AQI thresholds, pollutant categories and compliance factor for fuel under CPCB Guidelines and BSVI standards. The advanced machine learning algorithms are used to capture complex relationships between the input variables and the emission outputs, allowing to accurately predict and optimize. AQI monitoring within the proposed framework was performed independently of CO₂ emission prediction. Air Quality Index values were calculated using concentrations of major atmospheric pollutants including PM2.5, PM10, NO₂, SO₂, CO, and O₃ according to established CPCB guidelines. Since CO₂ is not included in AQI calculations, CO₂ prediction and AQI monitoring were treated as complementary but separate environmental indicators within the proposed framework. AQI values were obtained through the OpenWeatherMap API and interpreted according to CPCB air quality categories. Pollutant concentrations for PM₂.₅, PM₁₀, NO₂, SO₂, CO, and O₃ were collected and converted into AQI classifications following standard CPCB guidelines. AQI data were updated dynamically whenever new observations became available through the API service.

Moreover, the models developed are incorporated into an interactive decision support system built with that enables real time visualization and user interaction. The systematic methodology brings robustness, scalability, and practical application of the proposed solution to tackle challenges in terms of sustainable urban transportation and environmental impact.

3.1 Dataset description

Table 2 with description of dataset can be seen.

Table 2. Summary of datasets used in the proposed framework.

3.1.1 Canada CO₂ emission dataset

This dataset captures the details of how CO₂ emissions by a vehicle can vary with the different features. The dataset has been taken from Canada Government official open data website (Kaggle, n.d.-b). This is a compiled version. This contains data over a period of 7 years. The List of attributes in the CO₂ Emissions Canada dataset can be seen in Table 3 below.

Table 3. List of attributes in CO₂ emissions Canada dataset.

CO₂ Emissions Canada contains 7,385 real world vehicle data from 42 car makers, 2,053 models and 16 vehicle types such as compact vehicles, sedans, sport utility vehicles and trucks. Specific technical and fuel use information is included in each record, such as engine size (0.9 L through 8.4 L), number of cylinders (3 through 16), transmission type (27 different types), fuel type (gasoline, diesel, ethanol/E85, hybrids), and fuel efficiency (city, highway, and combined driving). Engine size (0.9 L to 8.4 L), cylinder count (3 to 16 cylinders), transmission type (27 variants), and fuel type (gasoline, diesel, ethanol/E85, and hybrids) are just a few of the technical and fuel related details that are included in each record. City, highway and combined fuel consumption and fuel efficiency (mpg) are provided. CO₂ emissions, the target variable, ranges between 96 g/km and 522 g/km with a mean of approximately 250 g/km, and is therefore composed of a diverse class of vehicles, with low and high emissions. The dataset’s richness allows to model nonlinear interactions between vehicle layout, fuel choice and emission consequences, enabling the development of machine learning models to forecast and optimise urban mobility scenarios powered by biofuels.

3.1.2 India CO₂ emission dataset

The “CO₂ Emissions India.csv” dataset (Kaggle, n.d.-a) contains six attributes: State, Year, Carbon_Emissions_MtCO Tons (MT) CO₂, GDP_BillionINR, Urbanization_Perc, and Energy_Use_TWh. Each record represents annual data for a specific Indian state, providing information on carbon emissions (MT CO₂), gross domestic product (billion INR), urbanization level (%), and energy consumption (TWh). The dataset enables state-wise longitudinal analysis of carbon emissions and their relationship with economic growth, urbanization, and energy use. By integrating environmental, economic, and demographic indicators, it supports multi factor assessment of sustainability challenges across different regions of India. The time series structure of the dataset further facilitates trend analysis, forecasting, and evaluation of the influence of urbanization and economic development on emission patterns over multiple years.

3.1.3 Data on CHETNA emissions

The second dataset (CHETNA, n.d.), which is more relevant to urban mobility and infrastructure optimization, emphasizes sector wise contributions to emissions, as opposed to regional aggregation as in the first dataset. Some of the variables are total emissions, power sector emissions, traffic emissions, residential emissions (scope 1 and scope 2), industrial emissions, aviation related emissions, and pollutant specific indicators such as NOx and PM2.5. By the presence of a date column, one can conduct fine grained temporal analysis, studying daily variations and patterns of emission sources. This data set is highly relevant for urban mobility and infrastructure planning as it gives an overview based on the contribution of the sectors to emissions, rather than an overview on the regional level as the first does. Fine grained temporal analysis is made possible by the inclusion of a date column, which makes it possible to examine daily variations and trends in emission sources.

The Canada CO₂ Emissions Dataset was selected because it provides comprehensive vehicle-level technical specifications together with experimentally measured CO₂ emission values, making it highly suitable for developing predictive machine learning models. Although the dataset originates from Canada, the relationships among engine size, fuel consumption, transmission characteristics, and CO₂ emissions are governed by fundamental engineering principles that are broadly applicable across different geographical regions (Javed et al., 2022). To improve contextual relevance, the Canadian vehicle dataset was integrated with Indian environmental datasets, including state-level carbon emission records, AQI measurements, and sector-wise urban emission statistics. Nevertheless, differences in driving behaviour, traffic congestion, road infrastructure, vehicle fleet composition, climatic conditions, and fuel quality between Canada and India may influence prediction accuracy. Therefore, the results should be interpreted as indicative of the proposed framework’s capability rather than as exact representations of Indian transportation conditions. Future research should incorporate large-scale Indian vehicle emission datasets to further improve model generalizability and applicability.

3.2 Data preprocessing

The first phase of the proposed methodology is data preprocessing as shown in Figure 2, which is vital for maintaining the reliability, consistency and quality of the datasets, before model development. The preprocessing pipeline starts with loading and integrating multiple datasets, after which the data quality is systematically evaluated. Values are missing and dealt with appropriately either imputation or removal are used, depending on the importance and distribution of the missing values. Categorical variables such as fuel type or transmission are encoded into numbers so they can be used by machine learning algorithms. Numerical features are then scaled or standardized, to ensure consistent ranges for features and ensure that the model is not biased toward higher values. Outliers that might affect the model’s performance are determined and addressed using capped or removal techniques. Then, following these procedures, the data is cleaned and well structured, ready for model evaluation and training. The preprocessing workflow includes data loading, missing value handling, categorical encoding, feature scaling, outlier detection, and outlier treatment, resulting in a cleaned dataset ready for machine learning modeling. This graphic is clear and reveals all the decisions taken for handling missing values, applying encoding techniques to categorical data, and using normalization and outlier handling techniques, all of which contribute to the robustness and suitability of data for machine learning applications. Outliers were identified using the Interquartile Range (IQR) method, as shown in Equations (1–3), where IQR = Q₃ Q₁), (Q₁) and (Q₃) representing the 25th and 75th percentiles, respectively. The lower and upper limits were calculated as Q₁ 1.5(IQR) and Q₃ + 1.5(IQR). Data points falling outside these limits were considered outliers. This preprocessing step helped improve data quality and reduced the influence of extreme values on model performance. Numerical attributes were imputed using median values, while categorical variables were imputed using mode values.

Outliers:

IQR = Q 3 − Q 1 (1)

Lower = Q 1 – 1.5 ( IQR ) (2)

Upper = Q3 + 1.5 ( IQR ) (3)

3.3 Feature engineering and machine learning models

In order to improve the model’s performance and extract valuable insights from the datasets, feature engineering was carried out. Several transformations and derived features were added to enhance the ability of the model to represent complex relationships between variables. Further, the emission intensity indicators like emissions per GDP (Gross Domestic Product) and emissions per energy consumption were calculated to understand the efficiency of respective regions and their environmental impacts. A “combined” fuel efficiency value was developed by combining city and highway fuel consumption values to represent more accurately a car’s performance under real driving conditions. A number of transformations and derived features were added to improve the model’s ability to capture the complex relationships between variables. Combined City and Highway fuel consumption data was used to create a combined fuel efficiency which provides a better representation of real driving conditions.

The combined fuel consumption was calculated using Equation (4) whereas the Emission intensity was calculated using Equations (5, 6):

FC combined = ( FC city + FC highway ) / 2 (4)

where FC_city represents city fuel consumption and FC_highway represents highway fuel consumption.

EmissionGDP = CO 2 GDP GDP (5)

EmissionEnergy = CO 2 EnergyUse (6)

The intensity of emissions in different regions was also calculated to better understand their effectiveness and impact on the environment, such as emissions per GDP (Gross Domestic Product) and emissions per energy consumption. Time based data were processed to identify temporal features (annual trends and seasonal patterns) to enable time series analysis and improve forecasting. Sector wise emissions have also been aggregated in order to allow comparison study across the various emissions sources, such as transportation, power generation, industrial process etc. These feature engineering techniques significantly enhanced the model’s ability to identify patterns and its accuracy in predictions. CO₂ prediction and selection of fuel optimization were carried out with the help of several machine learning algorithms. Linear Regression was initially used as a baseline model to gain a basic understanding of the linear relationships between input features and emission outputs. More complex and nonlinear interactions were captured by using ensemble based approaches such as Random Forest Regression, which is an ensemble of multiple decision trees, improving prediction accuracy (MDPI, n.d.; Zhang et al., 2023). Decision Tree Regression was also employed due to its interpretability and the fact that it showed the importance of features, which was useful to understand the influence of every feature. The Random Forest prediction serves as the baseline emission estimate. Fuel coefficients were applied to represent the relative emission characteristics of different fuel types, including E10, E20, B20, conventional fuels, and Bio CNG. AQI adjustment factors were introduced to account for environmental conditions and provide contextual environmental assessment. The final prediction therefore combines machine learning outputs with domain specific environmental knowledge to improve decision support capabilities. Table 4 shows the evolution framework of the ML models used.

Table 4. Machine learning model evaluation framework.

Model performance was evaluated using R², MAE, and MSE on identical training, validation, and testing datasets to ensure fair comparison. Further, Decision Tree Regression was chosen as it has the advantage of interpretability and feature importance highlighting, making it useful to understand the impact of individual variables. Gradient Boosting models (Rao et al., 2024) were also considered as a means to further improve predictive performance by sequential learning and error minimization. The performance of all these models was evaluated by using standard statistical measures including R² score (which measures the percent of variance explained by the model), Mean Absolute Error (which measures the average prediction error) and Mean Squared Error (which punishes larger errors). These evaluation metrics allowed to perform an accurate and reliable assessment of the model’s accuracy and reliability.

Model Validation: The dataset was divided into training (70%), validation (15%), and testing (15%) subsets to ensure reliable model development and unbiased performance evaluation. The training set was used to fit the machine learning models, while the validation set was employed for hyperparameter tuning and model selection. The independent testing set was reserved exclusively for the final evaluation of model performance. The models were assessed using the coefficient of determination (R²), Mean Absolute Error (MAE), and Mean Squared Error (MSE). The Random Forest model demonstrated consistently high performance across the validation and testing datasets, indicating good generalization capability and minimal evidence of overfitting. The ensemble learning mechanism of Random Forest, which combines predictions from multiple decision trees, further enhances model robustness by reducing variance and improving prediction stability. Final predictions were done on the bases of Equation (7).where FuelCoefficient represents the relative emission impact of different fuel types and AQIFactor represents environmental adjustment based on prevailing air quality conditions. Coefficient values were derived from published literature, government reports, CPCB guidelines, and biofuel performance studies. Calibration was performed through iterative comparison of predicted values against observed emission trends.

FinalPrediction = RFPrediction × FuelCoefficient × AQIFactor (7)

The final machine learning model utilized the following predictor variables: engine size, cylinder count, transmission type, fuel type, city fuel consumption, highway fuel consumption, combined fuel consumption, fuel efficiency, GDP, urbanization percentage, energy consumption, emissions per GDP, emissions per energy use, and selected temporal features derived from annual observations. These variables were selected based on domain relevance and their contribution to predictive performance.

3.4 System architecture and web based platform development

In order to combine machine learning predictions with real time visualization, an interactive decision support system was created using the framework. The system architecture is designed to be a layered system, with a data input layer, a machine learning model layer, and a visualization layer as depicted in Figure 3. The data input layer allows users to input vehicle specifications and seamlessly integrates multiple datasets for analysis. These are passed to the layer of ML model that acts as the prediction engine, generating emission estimates and optimization insights. The visualization layer enables accessibility and ease of interpretation through an interactive and user friendly interface of these outputs. The system has an interactive platform that presents emission statistics and model outputs in an easy to use way.

It is designed to support dynamic visualization of urban emission patterns through the incorporation of multiple modules, such as data visualization, a prediction engine, and real time input handling. Furthermore, graphical tools like trend lines and comparative charts are employed to depict emission trends by time and sector. The platform includes a predictive module to estimate emissions based on user defined vehicle parameters, state wise and sector wise comparison to facilitate regional and source based analysis and summary metrics displaying emission statistics.

To facilitate reproducibility, all preprocessing procedures, feature engineering methods, model training strategies, evaluation metrics, and data sources are explicitly documented. The use of publicly available datasets and widely adopted machine learning libraries enables future researchers to replicate the analysis, evaluate alternative algorithms, and compare their findings with the results reported in this study.

The proposed framework was developed using a set of widely adopted software tools and libraries. Python (version 3.x), developed by the Python Software Foundation under the PSF License, served as the primary programming language. Machine learning and data analysis were implemented using Scikit-Learn and Pandas, both maintained by their respective development communities and distributed under the BSD License. Numerical computations were supported by NumPy (BSD License), while Flask (BSD License) was used for developing the web-based application interface. Interactive data visualizations were created using Plotly, which is distributed under the MIT License by Plotly Technologies Inc. In addition, real-time environmental data were obtained through the OpenWeatherMap API, which operates under a commercial/API license provided by OpenWeather (Hunter, 2007).

4 Results and discussion

4.1 Result

The results section contains the analysis of the vehicular CO₂ emissions, the air quality monitoring and the performance of the machine learning models, and the environmental impact of the conventional and biofuel transportation systems. The proposed system is designed to combine machine learning algorithms, environment analysis, real time Air Quality Index (AQI) monitoring, and interactive visualization in order to provide a platform for intelligent transportation analysis for smart city applications. Exploratory Data Analysis (EDA), results of the system implementation, emission prediction performance, biofuel optimization assessment, AQI monitoring and sustainability evaluation are included in the study. Several graphical and statistical analyses were carried out to consider the correlations between the different vehicle parameters, such as engine size, number of cylinders, fuel type, gear ratio, fuel consumption and CO₂ emissions. These analyses were used to establish critical variables that affect vehicle emissions and insightful information on sustainable fuel solutions for minimizing environmental impacts. The developed system was assessed with a composite dataset consisting of the Canada CO₂ Emissions dataset and India specific environmental datasets including the records of AQI, State wise emission statistics and trends in urban transportation.

The machine learning model primarily aims at forecasting CO₂ emissions from vehicles, while the AQI model is used as a standalone system to monitor and visualize urban air quality conditions, using regular air pollutants. Therefore, CO₂ emission prediction and AQI assessment is regarded as complementary, but separate environmental indicators in the framework. Conventional fuels and alternative fuels (Ethanol E10, Ethanol E20, Biodiesel B20 and Bio CNG) have been compared, highlighting the potential environmental benefits of cleaner transportation fuels. In addition, the web based visualization platform offers real time information regarding emissions related to transportation and urban air quality in major cities of India.

The Random Forest model demonstrated superior predictive performance compared with the other evaluated algorithms owing to its ability to model complex nonlinear relationships among vehicular parameters and emission outputs. The obtained performance metrics indicate that ensemble learning approaches are more suitable for transportation emission prediction than traditional linear models. These findings are consistent with previous studies that reported improved prediction accuracy using Random Forest based approaches for environmental and transportation datasets.

4.2 Exploratory data analysis visualization

The vehicular emission dataset is then analysed using Exploratory Data Analysis (EDA) is presented in Figure 4, which includes multiple statistical and graphical visualizations of the data to understand the behaviour of CO₂ emissions and how they relate to various parameters of the vehicles. The visualization as a whole offers a thorough overview of the emission trends, the distribution of features, data quality, and relationships between the variables. The histogram of the CO₂ emissions distribution shows that most vehicles emit a moderate amount of CO₂, and a relatively few vehicles emit a very high amount. The distribution is slightly skewed to the right, suggesting the presence of a few outlier vehicles with high emissions, likely SUVs and pickup trucks. The data in the graph reveals that there are 7,385 observations in the data set, the mean CO₂ emission value is about 196.67 g/km. There is a wide range of emission behaviour between the vehicles, ranging from 71 g/km to 663 g/km. The boxplot analysis showing the range of emissions from various fuel types showed that the conventional fuel (petrol and diesel) has a wider range of emissions than the other cleaner and greener fuel (Chen et al., 2022) types (hybrid, CNG, E20 ethanol based). The median emission levels for hybrid and ethanol fuels are relatively low, and the distributions are narrower, indicating better fuel efficiency and lower environmental impact. As seen from the scatter graph there is a strong positive correlation between engine size and CO₂ emissions. Taller cars have a much greater impact on carbon emissions. The visualization features a correlation coefficient of ~0.72, which means that engine displacement is one of the biggest factors that contributes to the generation of vehicular emissions. The same can be said of the cylinder count analysis: cars with more cylinders generate more CO₂ as they require more fuel and engine power. According to the transmission type distribution chart, the transmission type distribution for the whole data set shows the number of automatic transmission vehicles has surpassed the number of manual transmission vehicles by over 50%. AMETRIC TONS (MT) and CVT systems are relatively small amounts of the data. Further box plots of the transmission systems show that, on average, the emission levels are slightly greater for the automatic transmissions than for the manual and CVT transmissions, but there is significant overlap in the emission ranges because of the effect of other vehicle parameters such as engine size, fuel consumption, etc. Interrelationships among important numerical variables, like engine size, number of cylinders, fuel consumption and CO₂ emissions, are strong as seen in the pair plot visualization.

There are linear relationships that are positive between engine size and fuel consumption and between fuel consumption and carbon emissions. The relationships are valid and illustrate that larger vehicles with higher fuel consumption have a more significant impact on environmental pollution. The vehicle class analysis also shows the environmental effects of various vehicle classes. However, SUVs and pickup trucks have the highest average CO₂ emissions, while compact cars, subcompact cars and two seater cars are relatively low in emissions (Hunter, 2007). This variation is primarily attributable to the differing engine capacities, vehicle weights and fuel economy attributes of each vehicle type. Another significant visualization displayed is the scatter plot of fuel consumption vs. CO₂ emissions. The graph indicates that the parameters have a strong positive correlation with the correlation coefficient of about 0.86 and R² of 0.74. This suggests that fuel consumption is a key factor in carbon emissions prediction modelling and one of the best indicators of vehicular carbon emissions. The nonexistence of missing values suggests that the data quality is good and the data set is reliable enough to be used for machine learning analysis without needing to resort to lengthy data imputation methods. Unless otherwise stated, figures presented in this section are generated using a combination of observed environmental data, machine learning predictions, and simulation based biofuel scenarios implemented within the proposed framework. Dashboard values represent analytical outputs derived from integrated datasets and should not be interpreted as direct measurements unless explicitly indicated.

Overall, the Exploratory Data Analysis (EDA) provides comprehensive insights into vehicular Shows the emission behaviour and emphasizes the significant impact of different vehicle related parameters on urban CO₂ emissions. It was found that engine size, number of cylinders, fuel consumption rate, transmission system, fuel type and vehicle category all have a great influence on emissions. The analysis revealed that vehicles with higher engine sizes and fuel use had significantly higher CO₂ emissions, and that vehicles with lower fuel consumption and cleaner fuel had relatively low CO₂ emissions. The dataset was analysed using a variety of visualization techniques, such as correlation analysis, distribution plots, and categorical comparisons, which were used to identify important trends, relationships, and patterns in the data. A series of visualization techniques were employed to gain a deeper understanding of the dynamics of urban transportation pollution, such as correlation analysis, distribution plots, and categorical comparisons.

The analysis helped to determine feature dependencies, data distributions, outliers and key predictive features for accurate forecasting of CO₂ emissions. It also confirmed the significance of cleaner fuels, sustainable transportation technologies and fuel efficient systems for vehicles in order to diminish environmental pollution and enhance urban air quality. The EDA helped to improve the quality of the data and facilitated the effective selection of features, which significantly aided in the creation of reliable predictive models and the overall reliability, interpretability, and effectiveness of the proposed ML based urban mobility and environmental monitoring framework.

4.3 Overview of the deployed biofuel powered urban mobility platform

The proposed ML Driven Biofuel Powered Urban Mobility Analysis system has been successfully implemented as an intelligent, interactive platform for analysing vehicular emissions, monitoring urban air quality, predicting CO₂ emissions, and suggesting sustainable biofuels alternatives for transportation systems in India. The proposed solution integrates machine learning algorithms, real time environmental data collection, predictive analytics, and interactive visualization tools to tackle the escalating problem of vehicle pollution and deteriorating air quality in urban areas of India. The development of the system was specifically designed with an India centric smart city approach (Kumar et al., 2023; Wang et al., 2020), focusing on urban regions such as Delhi, Mumbai, Bengaluru, Chennai, Hyderabad and Kolkata which are most impacted by transportation emissions and thus environmental degradation. The framework includes several databases for conducting a full environmental and vehicle analysis. The Canada CO₂ Emissions dataset was used to train the machine learning prediction model, which consists of over 7,000 vehicle records and detailed vehicular parameters like engine size, number of cylinders, transmission type, fuel type, and fuel consumption values. This dataset was chosen due to its organised data on automotive information and its applicability for emission prediction modelling. India specific datasets for state wise carbon emissions, urbanization, AQI conditions (Breiman, 2001; Gupta et al., 2024), and sector wise pollution data, were added to the web based platform and analytical modules to make the framework more relevant to the Indian context. The hybrid system enabled the integration of the predictive capabilities of the Canadian automotive dataset and the actual environmental conditions in India along with the smart city transportation issues.

The entire platform was developed, providing the flexibility to build an interactive and responsive web based platform (Biofuel, n.d.) with real time visualization and user interaction. Several machine learning algorithms were considered in the development process, but the final predictive engine was developed using the Random Forest Regressor for its ability to deal with nonlinear relationships, to make accurate predictions and to perform in a stable way under different vehicle conditions. The predictive framework also implements a rule based prediction methodology that uses engine characteristics, fuel type, and vehicle class to determine the relationships between them, where the relationships are weighted by domain specific coefficients determined from emission patterns. The developed web based platform named Biofuel Powered Urban Mobility (Biofuel, n.d.) is a combination of machine learning based emission prediction and real time monitoring of the surrounding environment, enhancing the interpretability of the results and assisting informed, data driven decision making for sustainable urban mobility and biofuel optimization. It has platforms with emission analysis, ML prediction, optimization of biofuel and AI assisted recommendations based on BS VI standards and the India’s E20 (2025) program. The created Biofuel Powered Urban Mobility platform, combines real time environmental monitoring with machine learning based emission prediction. The platform features dynamic graphics and state by state comparisons to help illustrate key metrics like total CO₂ emissions, vehicle emissions and potential biofuel savings. Additionally, it contains modules for biofuel optimization, ML prediction, emission analysis, and AI assisted recommendations that are in line with India’s E20 (2025) goal and BS VI regulations.

The platform is then composed of five integrated modules: Overview of the platform, Emission Analysis Module, ML Based Predictor, Biofuel Optimizer, and AI Assistant. These modules come together to create a smart city oriented environmental decision support system that can help researchers, policy makers, urban planners and students in understanding the transportation induced pollution and assessing the sustainable mobility solutions for Indian cities. The developed platform is shown in Figure 5 successfully brings together the machine learning based emission prediction, real time monitoring of AQI, environmental simulation, and biofuel optimization. A modern and professional web based platform layout with dynamic charts, metric cards and color coded indicators for environmental interpretation has been designed. The green indicators are for conditions that are environmentally friendly and have lower pollution levels, whereas the orange and red indicators show higher levels of pollution and emission.

Real time AQI monitoring (Rao et al., 2024) was achieved by integrating the API based environmental data retrieval from major Indian metropolitan cities. The AQI values are updated dynamically for the cities like Delhi, Mumbai, Bengaluru, Chennai, Hyderabad and Kolkata. The monitoring results showed that Delhi had one of the highest levels of AQI, and on days of heavy traffic, vehicle emissions, industrial pollution and atmospheric conditions, the AQI would fall under the “Unhealthy” category. The web based platform visualizations are effective in showing the impact of transportation related emissions on urban air quality in India. The coupling of AQI analysis and vehicle emission prediction model gives a realistic sense of the environmental condition in urban areas and enhances the smart city applicability of the proposed framework. To evaluate platform performance, the deployed system was tested under multiple user input scenarios. The platform demonstrated stable operation during repeated prediction requests and dynamic visualization updates. Average response times for prediction generation remained within a few seconds under normal operating conditions. The modular architecture supports future scalability through integration of additional datasets, predictive models, and environmental monitoring modules. Although formal user experience evaluation was not conducted, the platform successfully supported real time interaction, visualization, and decision support functionalities during testing.

To validate the functionality of the developed platform, multiple vehicle configurations representing different engine sizes, fuel types, and vehicle classes were tested through the web interface. The generated predictions were compared with expected emission trends reported in the literature and official emission datasets. The platform consistently produced realistic emission estimates, correctly identifying higher emissions for larger engines and conventional fuels while predicting lower emissions for biofuel based transportation scenarios. The successful operation of the prediction engine, visualization modules, AQI monitoring components, and biofuel optimization functions demonstrates the practical feasibility of the proposed framework for sustainable transportation analysis.

4.4 Emission analysis results

The Emission Analysis module in Figure 6 offers detailed, state and city wise emission analysis using interactive charts and graphical visualization. The module simulates the emission scenarios of traditional fossil fuel powered transportation with biofuel powered transportation scenarios under Indian urban conditions. The state wise emission comparison analysis showed that the states with high level of urbanization and industrialization had higher transportation related CO₂ emissions. However, the emission values were high in Maharashtra, Gujarat, Uttar Pradesh, Karnataka and Delhi, attributed to their dense transportation networks, high vehicle load and growing consumption of fuel. The analysis showed that ethanol blending of E20 in conventional fuel system resulted in a significant reduction in predicted CO₂ emissions in various states in India. This discovery aligns with the National Biofuel Policy and the Government of India’s E20 (Khatri and Kumar, 2024) ethanol blending program, which seeks to phase out the use of fossil fuels and cut carbon emissions from transportation. City wise emission distribution analysis also identified the major pollution hotspots in Indian metros.

Delhi and Mumbai accounted for a significant share of urban transport emissions because of the high traffic congestion and high vehicle density in these cities. Pie chart analysis allowed the contribution of the emissions to the city specific environmental burden to be visualized in a comparative manner, which facilitated the understanding of the city specific environmental burden. The contribution analysis by vehicle type revealed that passenger vehicles and commercial transportation systems are some of the biggest contributors to urban transportation emissions. SUVs, diesel powered vehicles and heavy transportation systems had higher predicted CO₂ emissions than compact vehicles and ethanol compatible transportation systems. The results highlight the importance of intelligent mobility planning, sustainable fuel options and environmentally optimized transportation in Indian smart cities.

4.5 Machine learning prediction results

The predictive framework was developed based on the Random Forest Regressor model, which is trained on the Canada CO₂ Emissions data set and optimized with India specific environmental and transport data set. The model is chosen due to its prediction accuracy, capability to model the nonlinearity of the emission behaviour, and being robust against overfitting. Major vehicle parameters such as engine size, number of cylinders, type of fuel, fuel consumption and vehicle class were used to train the Random Forest Regressor (MDPI, n.d.). In feature engineering and processing, these features were identified as the key aspects of the CO₂ emissions from vehicles. In addition to the machine learning model, the prediction system also follows a rule based predictive methodology calibrated using weighted domain specific coefficients. The cylinder count, fuel type and vehicle class contributed about 42, 18 and 12%, respectively, to the total emission prediction. This hybrid predictive approach enhanced the interpretability of the model, while retaining the prediction capability in different transportation conditions in India. Table 5 compares the predictive performance of the machine learning algorithms evaluated in this study. Random Forest Regression achieved the highest coefficient of determination (R²) and the lowest prediction errors (MAE and MSE), indicating superior capability in capturing nonlinear relationships between vehicle characteristics and CO₂ emissions. Linear Regression provided acceptable baseline performance but was unable to effectively model complex interactions among input variables. Ensemble based approaches, particularly Random Forest and Gradient Boosting, demonstrated improved prediction accuracy due to their ability to capture nonlinear patterns and reduce overfitting. Based on these results, Random Forest Regression was selected as the primary prediction engine of the proposed Biofuel Powered Urban Mobility Platform.

Table 5. Comparative performance of machine learning models.

Random Forest Regression achieved the highest predictive accuracy with an R² score of 0.970, the lowest Mean Absolute Error (MAE) of 4.07 g/km, and the lowest Mean Squared Error (MSE) of 52.83. Decision Tree Regression demonstrated moderate predictive capability, while Linear Regression and Gradient Boosting Regression showed comparatively lower performance. The superior results obtained by Random Forest can be attributed to its ensemble-learning structure, which effectively captures complex nonlinear relationships among engine characteristics, fuel consumption parameters, and vehicular CO₂ emissions. Therefore, Random Forest Regression was selected as the final prediction engine for deployment within the proposed BUMP platform. The superior predictive performance of the Random Forest model was consistently observed during validation and testing, demonstrating its ability to generalize effectively to unseen data. The small prediction errors obtained across the evaluation datasets indicate that the model did not exhibit significant overfitting. These findings confirm that the proposed machine learning framework is sufficiently robust for reliable CO₂ emission prediction under diverse vehicle operating conditions.

The ML Prediction module can be used to enter vehicle details and automatically predict CO₂ emissions in both conventional fuel and biofuel cases dynamically. Predicted emissions were around 185 g/km for a conventional petrol vehicle configuration (1.5 L 4 cylinder) in experimental simulations. The same car was also simulated with ethanol blend fuel E20 and the predicted emissions were lowered to nearly 145 g/km, which is an approximate 21% reduction. The prediction results clearly indicate that ethanol based biofuels can significantly reduce urban transportation emissions and it can be used in existing BS VI vehicles in India. The model worked well to produce stable and realistic forecasts in various modes of transportation, thus demonstrating that machine learning is applicable in smart mobility optimization and urban environmental analysis.

The Biofuel Optimizer module is designed to assess various biofuel options regarding the reduction of emissions, cost effectiveness, fuel availability and suitability to the Indian vehicle categories. The module is currently available for comparative analysis of Ethanol E20, Biodiesel B20, Bio CNG and Ethanol E10 systems. Optimization results showed that Bio CNG had the maximum estimated CO₂ reduction potential of around 23.8% with Ethanol E20 having the next highest emission reduction of nearly 21.5%. Biodiesel B20 achieved around 18.3% reduction and Ethanol E10 had moderate emission reduction potential at around 12.4%. Practical transportation aspects pertinent to India such as fuel affordability, cost impact, infrastructure availability and compatibility with BS VI vehicle systems were also taken into account in the analysis. Ethanol E20, being a low cost, high fuel availability and petrol based transportation system friendly solution, proved to be one of the most practical and scalable options for Indian smart cities. The optimizer classifies biofuels by type of vehicle such as petrol vehicles, diesel vehicles, heavy vehicles, flex fuel vehicles, and commercial transportation. This feature enhances the applicability of the framework for sustainable urban transportation planning in India. The results achieved are a strong indicator of the potential of India to shift toward cleaner mobility and also demonstrate the positive impact of large scale biofuel usage in urban smart city environments.

AI Assistant module was developed to enhance user interaction and offer intelligent guidance for the environment in the platform. The assistant was designed to be a smart recommendation system that can enhance the accessibility of the user and ease the analysis of the environment for students, researchers, and policy makers. Additionally, the chatbot interface showcases future proofing with the ability to connect with more sophisticated conversational AI solutions and large language models. The AI Assistant enriches the educational and smart city aspects of the framework through natural language interaction for environmental awareness and sustainable mobility analysis.

4.6 Smart City relevance and sustainability impact

In Figure 7 we can see the integrated Smart City Framework (Kumar et al., 2023; Sun et al., 2021) for Sustainable Urban Mobility and AQI Monitoring. Overall, the proposed ML driven biofuel optimization system is closely aligned with the goals of the Smart Cities Mission in India, which aims to use technology and innovation to create smart cities. Smart cities in India need adaptive and technology oriented urban infrastructure that can track the environmental conditions and decrease the pollution in the cities and make them more efficient in transportation. The proposed system helps to achieve these targets by providing real time vehicular emission analysis, environmental monitoring, optimization of biofuel and intelligent visualization of urban pollution patterns. This platform allows policymakers and urban planners to detect pollution hotspots, assess the environmental impact of transport and perform an analysis of the impact of alternative fuel strategies for Indian cities. The framework shows the potential of the machine learning in enabling cleaner transportation systems, lowering carbon emissions and enhancing urban air quality, using predictive analytics and simulation analysis. The incorporation of country specific environmental data, BS VI emission specifications, AQI monitoring, and E20 fuel optimization further enhances the applicability of the developed framework in the urban infrastructure system in India. The proposed platform exemplifies how technology can be leveraged to create a more sustainable, ecofriendly, and efficient transportation system for smart cities in India.

It also enables assessment of alternatives like Ethanol E20, Biodiesel B20 and Bio CNG to support the Indian Government’s sustainability efforts like the National Ethanol Blending Programme and BS VI emission standards. According to the simulation results, such biofuel transportation modes have the potential to substantially cut down CO₂ emissions from transportation as compared to conventional transportation modes based on fossil fuels, Although CO₂ itself is not included in AQI calculations, reductions in transportation related fuel consumption and associated pollutant emissions may indirectly contribute to improved urban air quality conditions. Therefore, CO₂ emission prediction and AQI monitoring are treated as complementary environmental indicators rather than directly linked variables. It provides urban planners, environmental agencies, policymakers and researchers with the ability to compare fuel strategies, assess sustainable mobility policies, and improve decision making with integrated state wise emission analysis, AQI monitoring, predictive analytics, and visualization of the environment. The proposed system represents a compelling approach to integrating AI, predictive modeling, and sustainable fuel optimization to improve environmental management, foster public health benefits, lower greenhouse gas emissions, and contribute to the creation of environmentally sustainable and smart cities in India.

4.7 Comparative analysis of conventional fuel and biofuel systems

The comparative analysis was conducted to assess the environmental impact of conventional fossil fuel transportation system and biofuel based transportation system under the condition of Indian urban surroundings. The analysis was based on the comparison of the predicted CO₂ emissions (Biofuel, n.d.; Chen et al., 2022; Shah et al., 2023) produced by petrol/diesel vehicles and the alternative fuel systems like Ethanol E20, Biodiesel B20, Bio CNG, and Ethanol E10. The comparison was performed with machine learning based simulations that are embedded in the created platform framework and is shown in Figure 8. The results obtained clearly showed that the transportation system with conventional vehicles (based on fossil fuels) consistently emitted more CO₂ than the transportation system with biofuels. Insufficient combustion of petrol and diesel vehicles, traffic jams and higher consumption rates lead to high emissions in dense urban traffic. The overall reduction in predicted emissions was the highest for Bio CNG and closely followed by Ethanol E20 fuel systems of all the analysed biofuel alternatives.

The adoption of E20 fuel blends demonstrated particularly promising results for India, given that E20 blends can be used with existing BS VI petrol vehicle infrastructure and the Government of India’s current ethanol blending programs. The vehicles tested under E20 conditions consistently had reduced predicted CO₂ emissions without compromising on operational stability or fuel efficiency. The comparative analysis also showed that for compact and medium sized vehicles, ethanol blending was found to be beneficial while for heavy commercial vehicles, optimization under Bio CNG and biodiesel system was found to be beneficial. The results underscore the need to consider fuel strategies on a vehicle by vehicle and transportation by transportation basis in urban settings. This developed web based platform effectively visualized these comparisons, with interactive bar graphs, pie charts and emission trend visualizations, providing an easy to understand view of the environmental differences between conventional and alternative fuel systems. The findings are very strong in favour of using biofuels for transportation as a part of India’s smart city and sustainable mobility projects. The relative ranking of Bio CNG, E20, and B20 observed in this study is broadly consistent with previous transportation emission studies that reported significant greenhouse gas mitigation potential for biomethane and ethanol based fuel blends. The findings suggest that alternative fuel selection can substantially influence transportation related environmental performance and support national decarbonization objectives.

4.8 Performance evaluation of the machine learning model

The machine learning framework performance was validated using various scenarios of prediction and simulation with varying vehicle configurations, fuel types and operating conditions as shown in Figure 9. High predictive ability was achieved by the Random Forest Regressor because of the structure of ensemble learning and because the model is able to simulate the nonlinear emission behaviour for various transportation conditions. The model was able to learn relations between engine size, fuel type, number of cylinders and vehicular emission output. The size of the engine and number of cylinders were consistently related to higher predicted CO2 emissions. Smaller vehicles and ethanol fueled systems resulted in lower values.

The results validated the effectiveness of feature engineering and preprocessing used to create the model. In the Figure 10, We can see Random Forest model also showed stable performance in various vehicle categories and fuel combinations. The predictions for emissions were similar across various testing conditions, demonstrating good generalization and robustness to data variability (Zhang et al., 2023; Wang et al., 2020). The machine learning system was particularly effective in the deployed platform for generating predictions in real time. Users can change the specification of a vehicle during use and immediately view how these predicted emissions change with the varying fuel conditions. User friendliness and applicability of the proposed framework is increased by the interactive prediction feature, which makes it a valuable tool in the context of educational, research and smart city planning applications. The outcomes from the tests indicate that machine learning based environmental prediction systems can assist in smart transportation management and sustainable mobility optimization in the fast urbanizing areas.

4.8.1 Statistical validation of model performance

Fivefold cross validation was performed to evaluate model robustness. The Random Forest model demonstrated stable performance across all folds with minimal variation in prediction accuracy, indicating good generalization capability and low risk of overfitting.

4.9 Real time AQI and environmental impact analysis

The developed system with real time AQI monitoring (Rao et al., 2024) aided in detailed analysis of relationship between transportation emissions and urban air quality in Indian cities. The study relied on the continuous analysis of AQI by integration with API and transport emission simulations to understand the trends in the environment of metropolises. The analysis revealed that the places with a high level of traffic congestion had a consistent decline in AQI. The frequent AQI values of Delhi observed under the “Poor,” “Very Poor,” and “Unhealthy” conditions, emphasize the serious impact of transportation emissions on public health and environmental sustainability. The high AQI levels were seen in Mumbai, Bengaluru and Kolkata at their peak traffic time as well. It is important to note that CO₂ emissions and AQI represent two different environmental indicators. AQI is calculated using concentrations of pollutants such as PM₂.₅, PM₁₀, NO₂, SO₂, CO, and O₃ and does not directly include atmospheric CO₂ concentrations. Therefore, within the proposed framework, AQI monitoring and CO₂ emission prediction were treated as complementary but independent analytical modules. The reduction in predicted CO₂ emissions under biofuel scenarios indicates improved carbon performance, while AQI monitoring provides information regarding ambient air quality conditions and potential public health impacts.

The results of the developed platform simulations showed that the use of biofuel powered transportation systems would improve significantly urban air quality by reducing carbon emissions and associated pollutants from vehicles. Pollution intensity was found to be lower for ethanol based fuels than for conventional petrol and diesel systems, and the pollution levels were lower for simulated smart city conditions. AQI monitoring (Breiman, 2001; Gupta et al., 2024) and machine learning prediction are also integrated into an environment friendly management system that facilitates smart urban planning. Such systems can be used to help pinpoint pollution hotspots; track transportation related environmental risks and measure the effects of policies for cleaner fuels. The developed framework illustrates how the real time environmental analytics and machine learning technologies can play a role in achieving a cleaner and healthier urban environment in the context of smart city infrastructure.

5 Discussion

The present study successfully shows that the integration of machine learning, biofuel optimization and environmental analytics in one end to end system for sustainable urban transportation management is possible. The developed ML Driven Biofuel Powered Urban Mobility Analysis System has successfully integrated the concepts of predictive modelling, AQI monitoring, emission analysis, and interactive visualization to create a comprehensive solution for smart cities in India tackling the escalating issue of urban pollution generated by vehicles. The results achieved in this work have shown that machine learning (MDPI, n.d.; Zhang et al., 2023) techniques can be very effective in predicting and analysing transportation related CO₂ emissions and can facilitate intelligent environmental decision making. The Exploratory Data Analysis showed that there were good relations between the emission behaviour and the characteristics of the vehicles. The CO₂ emission generation was found to be strongly influenced by the features like engine size, number of cylinders, fuel consumption, and fuel type. The results showed a clear correlation between engine size and fuel consumption with the magnitude of emission values, with larger engines and higher fuel consumption consistently yielding higher emission values. The results obtained in this study are consistent with previous investigations reporting the effectiveness of Random Forest models for emission prediction and the environmental benefits of alternative fuels. Similar studies have demonstrated that ensemble learning approaches outperform traditional regression techniques in capturing nonlinear relationships among vehicle characteristics and emission outputs. Furthermore, previous investigations on biodiesel and biomethane powered transportation systems have reported measurable reductions in transportation related emissions, supporting the findings of the present study. The observed benefits of E20, B20, and Bio CNG therefore align with the broader literature on sustainable transportation and alternative fuel adoption.

The Prediction Results further verified that the biofuel based systems like Ethanol E20 and Bio CNG are considerably reducing the vehicular emissions as compared to the conventional fossil fuel based systems. The comparison of conventional fuels with biofuels gave valuable insights into the environment of sustainable transportation in India. The fuel system that performed best is Bio CNG followed closely by Ethanol E20 fuel systems. The E20 ethanol blend proved to be one of the most viable and scalable options given its current infrastructure of petrol vehicles in the country, especially BS VI vehicles, and the ongoing development of the ethanol infrastructure. The findings presented overwhelming evidence in favour of the Government of India E20 blending policy (Khatri and Kumar, 2024) and suggest a significant reduction of carbon emissions from transportation fuel blends when using biofuels on a large scale can improve urban air quality. The real time monitoring of AQI has made the developed framework very useful in practice. The web based platform successfully highlighted the direct impact the transportation sector is having on the air quality scenario of metros like Delhi, Mumbai, Bengaluru, and Kolkata. The main reason of unhealthy AQI in Delhi was the dense traffic flow, industrial activities and its high level of vehicle emissions. Based on the simulation results, the optimized implementation of biofuel powered transportation systems can be expected to positively affect AQI conditions through reduction in pollutant intensity and transportation related carbon emissions. The creation of the smart city web based platform with interface, which combines machine learning predictions, environmental monitoring, and fuel analysis for sustainable solutions within a single platform, is another significant aspect of this study.

The platform converts complex environmental data into an understandable and accessible user interface that can be used by students, researchers, urban planners and policymakers. The incorporation of AI driven suggestions and live visualization enhances the usability of the system, highlighting the system’s potential for the future of intelligent transportation management. Though the proposed framework has been successful, some shortcomings are still there. The main data set used to train the machine learning model was the Canada CO₂ Emissions dataset because of the availability of structured records of vehicles. While having produced India specific datasets that were integrated to predict AQI and analyse the environment, future studies may benefit from larger datasets of vehicular emission, which include information on India’s real world traffic and vehicular driving behaviour. Moreover, only CO₂ emissions and trends of AQI (Breiman, 2001; Gupta et al., 2024) have been taken into consideration in the current system; the introduction of more pollutants like NOx, SO₂, and particulate matter would further enhance the accuracy of environmental analysis. In general, the findings discussed validate the potential of ML based environmental management systems to revolutionize sustainable transportation planning, urban pollution monitoring and smart city infrastructure. The convergence of predictive analytics, optimizing biofuels, and intelligent visualization offers a scalable and flexible approach to help drive cleaner, greener, and more resilient urban mobility systems in India. The use of publicly available datasets, documented preprocessing procedures, defined feature engineering equations, and widely available machine learning libraries enhances the reproducibility of the proposed framework. Researchers can replicate the workflow using similar datasets and compare alternative modeling approaches against the baseline results presented in this study. The present study has some drawbacks. The training of the vehicle level models was done using a Canadian emission dataset, which may not be fully representative of Indian driving conditions and vehicle fleet characteristics. Second, the monitoring of AQI was based on easily available environmental data and not on dedicated IoT based sensor networks. Thirdly, the framework was mainly directed to transportation related emissions and excluded all possible urban emissions sources. Future research may involve using larger India specific datasets, integrating real time sensors, and optimizing the use of sophisticated deep learning models to enhance prediction accuracy and scalability.

The findings obtained in this study are consistent with previous research on machine learning based emission prediction and sustainable transportation systems. Similar to the work of Breiman (MDPI, n.d.) and Chen et al. (Zhang et al., 2023), Random Forest Regression demonstrated strong predictive capability for modeling nonlinear environmental datasets. The observed reduction in emissions under E20 and Bio CNG scenarios is also in agreement with studies reporting the environmental benefits of biofuel adoption in transportation systems (Biofuel, n.d.; Chen et al., 2022). However, unlike most previous studies that focus on either emission prediction, AQI forecasting, or biofuel assessment individually, the proposed framework integrates these components within a unified decision support platform. This integration enables simultaneous evaluation of vehicular emissions, environmental conditions, and sustainable fuel alternatives, thereby extending the practical applicability of machine learning based environmental management systems for smart city transportation planning.

6 Limitations of the study

Although the proposed framework demonstrated excellent predictive performance, several limitations should be acknowledged. The vehicle-level prediction model was developed using the publicly available Canada CO₂ Emissions Dataset because of its comprehensive vehicle specifications and measured emission values. While the underlying engineering relationships governing fuel consumption and CO₂ emissions are generally transferable, regional differences in vehicle fleet composition, driving behaviour, traffic congestion, climatic conditions, and road infrastructure may affect prediction accuracy when applied to Indian transportation systems. Furthermore, the biofuel scenarios evaluated in this study were based on simulation using published emission coefficients rather than large-scale field experiments. Future work will focus on incorporating Indian vehicle emission datasets, real-world biofuel performance measurements, and additional environmental variables to further improve the robustness and applicability of the proposed framework.

7 Conclusion

This study introduced the Biofuel Powered Urban Mobility Platform (BUMP), an integrated framework based on machine learning for CO₂ emission forecasting in vehicles, air quality measurement, biofuel evaluation, and sustainable urban mobility analysis. The platform consists of an integrated decision supporting system for smart city applications, which includes the environmental monitoring, the predictive analytics, the optimization of biofuel use, and the interactive visualization. A Random Forest Regression model was used to estimate CO₂ emissions of vehicles based on the following features: engine size, number of cylinders, fuel consumption, transmission type and fuel category. The model showed consistent prediction accuracy for various vehicle types and fuel categories, highlighting the potential of machine learning techniques in transportation emission studies. Among all the evaluated biofuels, Bio CNG and Ethanol E20 offered the maximum emission reduction potential with existing transportation infrastructure. The results provide further evidence of the progress India is making in transitioning to greener transport and the positive impact of sustainable fuels on the environment.

The proposed Biofuel Powered Urban Mobility Platform demonstrates the potential of integrating machine learning, biofuel optimization, environmental monitoring, and interactive visualization within a unified decision-support framework for sustainable urban transportation. The Random Forest model achieved excellent predictive performance, indicating its suitability for accurately estimating vehicular CO₂ emissions under varying operating conditions. Simulation results further demonstrated the environmental benefits of alternative fuels such as E20 ethanol, B20 biodiesel, and Bio-CNG, highlighting their potential to reduce transportation-related carbon emissions. Although the proposed framework integrates Indian environmental datasets with detailed vehicle-level emission data, the predictive model was primarily trained using a Canadian vehicle emissions dataset. Consequently, differences in driving behaviour, traffic conditions, fleet composition, and road infrastructure between Canada and India may influence model transferability. Future research will focus on incorporating large-scale Indian vehicle emission datasets, expanding real-time IoT integration, and evaluating additional machine learning algorithms to further improve model accuracy, scalability, and practical deployment in smart city transportation systems.

Future research will involve integration of real time IoT based environment sensors, further development of the framework with large scale transportation data from India, embedding advanced deep learning models to predict traffic and emissions, and adding electric vehicle analytics, intelligent traffic management, and multi city sustainability assessment to the platform. These changes will further enable the utility of the framework for supporting national and global sustainability goals and for further development of cleaner, smarter and more sustainable urban transportation systems.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Roberto Alonso González-Lezcano, CEU San Pablo University, Spain

Reviewed by: Saša T. Milojević, Faculty of Engineering University of Kragujevac, Serbia

Fahad Waqas Mir, Wuhan University of Technology, China