A hybrid MCDM approach for site selection of electric vehicle battery swapping station

1 Introduction

India has a broad and multifaceted transportation sector, with 146,195 km of National Highways, which is the second longest and busiest in the world (Ministry of Road Transport and Highways, 2024). Road transport in India contributes approximately 12% of the nation’s energy-related CO₂ emissions and is a powerhouse of air pollution in urban areas (International Energy Agency, 2023). A change in the road mobility sector can resolve the degradation of the environment, mainly influenced by motor vehicle emissions. Therefore, eco-conscious Electric Vehicles (EVs) are adopted globally to reduce pollutant emissions from automobiles and promote sustainable road transportation. The global EV-Thirty campaign, supported by the Indian government with the main objective to have at least 30 % of new vehicle sales be electric vehicles by 2030 (Lewis, 2024). To facilitate EV adoption, the Indian government has initiated public-private partnerships and private players to develop urban and state-level infrastructure. Although the government is taking numerous initiatives, the charging infrastructure has not reached the desired level.

The development of EV charging infrastructure is intrinsically linked to energy suppliers. Widely used energy suppliers for EVs are EV public charging stations; an absence of adequate EV public charging stations at strategic intervals, just like traditional fuel pumps, is one of the reasons why potential users are unwilling to adopt EVs (Hidrue et al., 2011). The prolonged charging time leads consumers to stay longer at the station, leading to inconvenience for EV users (Egbue and Long, 2012; Koirala et al., 2022). A lack of charging infrastructure and waiting time are significant limitations for EV adoption (Franke et al., 2012).

The huge population and rising urbanisation in the country are resulting in demand for EV charging infrastructure.

Commissioning of battery swapping station (BSS) provides solutions to these challenges; a batch of fully charged batteries is readily available, and the EV users only need to swap the depleted battery with the fully charged battery, which can be done automatically or manually, saving the waiting time of the end user (Sarker et al., 2015). Therefore, BSS stands as a practical and feasible solution with significant promise for EV adoption (similar to domestic gas cylinders) (Sarker et al., 2013; Hussain et al., 2024). Setting up BSS plays an important role in the effective operation, subsequently leading to EV adoption. The Parivahan report confirms that there is a rise in EV adoption in CMA since 2020 at a rate of 20.21% over the five years until 2024, which is a prominent factor for this study (Vahan Sewa, 2025).

The steady shift in transition from conventional vehicles to EVs needs a stable charging infrastructure in CMA. The CMA is situated on the southeastern coast of India, in the state of Tamil Nadu, comprising Chennai city, along with a portion of Kancheepuram, Tiruvallur, and Chengalpattu districts. It encompasses the urban jurisdiction of Greater Chennai Corporation (GCC) and its surrounding urban and peri-urban regions (Chennai Metropolitan Development Authority, 2013). Planning, development, and regulation are overseen by the Chennai Metropolitan Development Authority (CMDA). The CMA ranks among India’s most densely populated urban regions (Chennai Metropolitan Development Authority, G. of T. I, 2025), and its population is consistently increasing due to better economic opportunities within the region (Mathivanan and Duraisekaran, 2025). It was a major economic centre in India and incorporated various industries, including automotive manufacturing, information technology and healthcare.

Tamil Nadu has 1,524 public charging stations installed over the last five years, with Chennai accounting for 218 (PM E-Drive, 2025). The EV BSS infrastructure in CMA remains insufficient to support the increasing EV user base (Singh and Dubey, 2022). The rapid rise in EV adoption, juxtaposed with infrastructural limitations, has led to CMA being chosen as the study area to investigate effective solutions for urban mobility and charging infrastructure enhancement. This research aims to support CMA in overcoming its infrastructure challenges heightened by rapid urbanisation and ensure that the benefits of electric mobility can be fully realised through adequate and accessible support systems. The site selection problem is a decision-making problem that can be solved using the MCDM approach, which is considered a reliable technique among researchers focusing on site selection. Researchers have selected an optimum location for EV charging stations using the MCDM approach (Meng et al., 2013; Guo and Zhao, 2015; Kaya et al., 2020). In many studies, MCDM-Geographic Information System (GIS) - frameworks, such as AHP-GIS, FAHP-GIS, FTOPSIS-GIS (Erbaş et al., 2018; Ali et al., 2021; Gu et al., 2025; Mandal and Mondal, 2025) have been adopted to identify a suitable location by incorporating spatial data and expert judgment, which may not fully capture evolving user preference and demand patterns. Similarly, AI-driven approaches employing machine learning algorithms, including random forest and support vector regression, have strong predictive capability, but their complex structures often reduce transparency for policymakers and planners (Bojer et al., 2023; Yaghoubi et al., 2024). To overcome these limitations, this work aims to combine Criteria Importance Through Intercriteria Correlation (CRITIC) for objective weight derivation with both Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) and COPRAS as parallel ranking engines, allowing validation of results and reducing method-induced bias in site selection.

The site selection criteria for EV BSS are based on socio-economic and demographic criteria such as proximity to key locations. These factors are often considered in urban infrastructure studies related to site selection (Roy and Law, 2022; Krishankumar and Ecer, 2024; Maghfiroh and Kavirathna, 2023). Many studies highlight the significant role of economic parameters in infrastructure planning, like cost-relevance for land acquisition, which is widely recognised as a key metric for investor decision-making for EV BSS installation (Liu et al., 2024; Alici, 2025). The five criteria are constructed based on the perspective of the stakeholder (C1, C3) and consumer (C2, C4 and C5) by referring to existing literature. The data presented in Table 1 demonstrate the hybrid MCDM techniques that are employed by researchers in their previous works.

Table 1. Hybrid MCDM for site selection of EV charging station.

In EV charging infrastructure, demographic characteristics are the most important criterion. The necessity for charging infrastructure is greater in locations where electric vehicles are widely used, and highly populated places have higher demands for BSS services (Guler and Yomralioglu, 2020). In EV BSS site selection, a charging infrastructure mainly considers ease of logistical convenience and setup; it can be a site near a commercial or residential zone (C2) (Chen et al., 2022). It measures how close sites are to where people live, and the coordinates of residential clusters are used to ensure accessibility and population coverage. Land cost (C3) refers to the price of the land for the construction of a BSS workplace, storage area for depleted and fully charged batteries, and BSS vehicle parking purposes. The BSS has a continuous process of battery swapping, charging depleted batteries, and storing charged batteries. The fire hazard associated with an electrical fault or other operational issues cannot be ignored. The personnel involved in BSS should also feel safe enough to carry out their duties. Therefore, the proximity to emergency services criterion (C4) calculates the closeness to nearby fire stations at the selected location and assesses emergency response potential to handle unforeseen events. To evaluate service integration and convenience for EV users, proximity to public facilities (C5) is measured by road-related attributes, depending on the vehicle mobility (Mazza et al., 2024).

The adoption rate (C1) is directly linked to public awareness, which influences the success of EV infrastructure deployment, were included in this research. Bhat et al. (2022) studied factors related to the adoption of technology to understand how different factors affect the adoption behaviour of individuals. Table 2 represents the set of criteria from the literature which are selected for the study. Rather than treating demand as an assumption, this framework operationalises EV adoption rate as a quantifiable decision criterion fed directly into the MCDM model. This demand-side grounding is absent from infrastructure-focused prior work.

Table 2. Decision criteria for BSS location selection.

The main objective of the study is to establish an evaluation criteria model for BSS establishment and derive the weightage for each criterion using the CRITIC method. To determine the rank of selected locations in CMA by using the TOPSIS and COPRAS methods. Achieving low-carbon and sustainable mobility is important to ensure sustainable urban infrastructure; the current urban charging infrastructure planning framework does not consider the factors that encourage residents to choose sustainable transport modes, and existing EVBSS studies typically apply a subjective MCDM method. Despite significant advances in GIS-MCDM and AI-driven optimisation for EV infrastructure planning, limited research has focused on BSS location selection using an integrated objective-weight and dual-ranking framework.

The framework shifts the evaluation criteria from operational metrics toward user experience indicators, such as spatial convenience and service accessibility (proximity to residential areas, public facilities, emergency services). This positions consumers as active stakeholders rather than passive end-users in the planning process. Consequently, the proposed approach offers transparent, computationally efficient and practically applicable decision-support tools for sustainable EVBSS planning.

The upcoming sections are divided as follows: Section 2 describes the data collection. Section 3 outlines the methodology, including the selection of alternative locations. Section 4 presents the calculations performed and the corresponding results obtained from the analysis. In Section 5, the findings are analysed and discussed in detail. Section 6 highlights the limitations encountered during the study. Finally, Section 7 concludes the paper by offering recommendations based on the research outcomes.

2 Data collection

After determining the criteria for location selection, the next phase of the research involved gathering data. Based on the factors identified in Section 1, data were collected for the following variables for each selected alternative in the study: EV adoption rate, Proximity to Residential Area, Land cost, Proximity to emergency services, and Proximity to public facilities. The adoption rate of EVs is a primary input in determining the significant location and placement of battery swapping stations. It reflects the behavioural change in the population; it captures EV acceptance among potential users, which is vital in EV charging infrastructure and urban mobility planning. It measures how widely electric vehicles (EVs) are being adopted within a specific population or region. It is calculated using the formula:

EV Adoption Rate = ( Average of Registered EV ( 5 Years ) Average Total Registered Vehicles ( 5 Years ) ) × 100

Where the average of registered EVs refers to the number of people (or vehicles) who adopted EVs in a selected location for five years. The average total number of registered vehicles over five years represents long-term vehicle growth and demand patterns within the study area. Five years of vehicle registration data for each alternative location from the Tamil Nadu Regional Transport Office (RTO), categorised by vehicle type and fuel type, have been collected using the Parivahan dashboard, which is an online data analytics portal that collects data from RTOs across India (Vahan Sewa, 2025). By averaging the yearly data, fuel-wise vehicle trends in the region are identified.

This average includes categories such as four-wheeler (invalid carriage): passenger car (including special purpose vehicles for persons with disabilities). Heavy motor vehicles such as heavy goods vehicles (HGVs) and heavy passenger vehicles (e.g., large trucks and buses). Light motor vehicle, which includes light goods vehicle (LGV) and light passenger vehicle examples, such as delivery vans, small buses. Medium goods vehicle, medium motor vehicle, medium passenger vehicle and other than those mentioned above. The categories, including non-transport three-wheelers, transport three-wheelers, special-purpose two-wheelers, and non-transport two-wheelers, are standard and capture the share of each fuel type- diesel/hybrid, electric, petrol/hybrid, plug-in hybrid EV, pure EV, strong hybrid EV registered over the years. It highlights how electric vehicle registrations have evolved in the alternative location relative to conventional fuels.

To calculate average land cost, the government-guided values for each street in an alternative location have been taken (Registration Department - Tamil Nadu, 2025). The calculated average land value ensures standardisation and transparency in land valuation. Using government-guided values and averaging multiple data points provides a realistic and fair basis for estimating land cost in the location selection.

The government hospitals, colleges and transportation terminals are taken as public facilities for this study. To identify how close a proposed candidate site is to residential areas, fire stations, or public facilities, Google Apps Script is used to calculate proximity between two addresses based on real-world road data from Google Maps (Keawthong et al., 2022). Three inputs have been used they are the origin address, as residential areas, fire stations, or public facilities; the destination address, as the proposed location; and the desired return type, as “kilometres.” a custom Google Sheets function called GOOGLEMAPS(), which calculates the distance. This approach provides an automated and realistic way to measure proximity for planning EV BSS location selection.

3 Methodology

To facilitate EV adoption in CMA by increasing the charging infrastructure, it is necessary to make a preferential selection of significant locations for setting up the EV battery swapping stations. The present study intentionally focused on five high-impact and measurable criteria to maintain methodological clarity and avoid excessive model complexity during initial framework development. Each criterion was selected on two grounds: first, it addresses a demonstrably important dimension of BSS site suitability (demand, feasibility, accessibility and cost); and second, using data sources consistently available within the study context.

In this research, the ideal location is typically determined by using the hybrid MCDM technique. The CRITIC method has been utilised to calculate the criteria weight without relying on any pre-existing knowledge of the experts or judgment from the stakeholders (Hassan et al., 2023). This technique is notable for removing all possible bias related to subjective assessment, thereby increasing objectivity. It evaluates the intensity of distinction as well as the inter-criterion discord held between the criteria (Zhao and Liu, 2025).

The TOPSIS and COPRAS techniques have been employed to assess and prioritise the selected alternative site alignment with the selection criteria for EV BSS development. To evaluate the uncertainties in the framework, Monte Carlo simulations were performed by perturbing the criteria weights in this study (Motlagh and Li, 2025). This simulation investigates how changes in input factors, such as randomly fluctuating the objective weights generated by CRITIC, affect the final result on TOPSIS ranking and serves as a rigorous tool to verify the robustness of the framework.

3.1 Location selection

The CMA is one of the urban zones with the highest population density (8,653,521) in India (Chennai Metropolitan Development Authority, G. of T. I, 2025). The CMA lies within the five administrative divisions of Tamil Nadu, viz. Chennai District, taluks and blocks of Thiruvallur District, parts of Kancheepuram District, areas of Chengalpattu District, and parts of Ranipet District. In this paper, six locations from CMA have been selected, namely Ambattur (L1), Tiruvallur (L2), Poonamalle (L3), Tambaram (L4), Sriperumbudur (L5), and Chengalpattu (L6), which are marked in Figure 1.

Ambattur (L1) is a Taluka in Thiruvallur District of Tamil Nadu, which stands out with the highest total population of 9,06,580 and urban population of 8,52,495. Chengalpattu (L6) has the largest total area among the selected alternative locations, which is 778 km² and the largest rural area, 624.15 km², as well as the highest rural population, 2,64,378. Although Chengalpattu (L6) has the largest urban area, Ambattur (L1) has the largest urban population. Tambaram (L4) has the highest population density. Sriperumbudur (L5) has a significant balance of urban and rural populations and a total area of 642 km².

The selected sites from the CMA provide a suburban and peri-urban dynamics which allow for a more heterogeneous analysis. The nearby supporting facilities for each of the six selected locations are relevant for EVBSS site selection, with high vehicle density and easy access from major roads or transportation hubs, ensuring consistent demand and convenience for users. The largest dedicated Ambit IT Park in Ambattur hosts numerous technology companies.

Key business parks like IndoSpace Industrial Park and Motherson Industrial Park in Tiruvallur are home to logistics and warehousing operations. A major software park, RMZ, located on Mount Poonamalle Road, serves several technology companies. Significant daily inflow of IT professionals supports EV adoption. Cognizant and the corporate office in Madras Export Processing Zone near Tambaram are drawing many employees and facilitating viable EV infrastructure usage. A State Industries Promotion Corporation of Tamil Nadu Limited (SIPCOT) Industrial Park, Sriperumbudur, and a modern IT park like Estancia IT Park at the gateway to Chengalpattu, housing several prominent software companies, attract a substantial technical workforce and commuting traffic, ideal for battery swapping and other EV services.

The importance of the criteria is determined based on data through the CRITIC method, which eliminates subjectivity in expert judgments in this study (Wei et al., 2020). Studies approached GIS with various MCDM methods, such as ENTROPY, AHP and TOPSIS, to select a suitable site (Spyridonidou et al., 2022). Many optimum site selection studies related to energy are obtained using TOPSIS and the CRITIC method to rank the alternative sites and calculate the criteria weights (Hassan et al., 2023; Taherdoost and Madanchian, 2024). In this study, the CRITIC method has been used to form the decision matrix, with the aid of various publicly available data sources, which are presented in Table 3. based on objective values, TOPSIS and COPRAS methods were used to rank the alternatives. A summary of the applied methods is described in the following sessions.

Table 3. Calculated values for the criteria.

3.2 CRITIC method for determining criteria weights

The CRITIC method is employed in the study to identify criteria which have high discriminating power and unique information, which enhance the robustness and reliability of the decision by computing unbiased objective weights for criteria without involving subjective judgment (Diakoulaki et al., 1995). The following are the steps for computing the criteria weights.

Step 1: The decision matrix is formed using the values from data sets, and all alternatives have been measured based on their performance values.

Step 2: The transformations of performance value x ij T are calculated using the following formula (Equation 1).

x ij T = { x ij − x j − x j ∗ − x j − if j ∈ B ; x j − − x ij x j − − x j ∗ if j ∈ N ; (1)

x j = ( x 1 j T , x 2 j T , … x nj T )

Where ( x ij ) is the original value for alternative i and criterion j. ( x j − ) is the minimum value of criterion j and ( x j ∗ ) is the maximum value of criterion j, respectively. B is a beneficial criterion where maximum is best, and minimum is worst. N is a non-beneficial criterion where minimum is best, and maximum is worst.

Step 3: Each criterion’s standard deviation is derived based on its respective vector.

Step 4: The m × m square matrix R is constructed (Equation 2).

R = [ r jk ] m × m (2)

The elements of R represent the correlation coefficients measuring the linear relationship between the criteria x j and x k. where r jk signifies the correlation among the criteria.

Step 5: The information measure of each criterion is calculated using the following formulae (Equation 3).

H j = σ j ∑ k = 1 m ( 1 − r jk ) (3)

Step 6: Now the criteria weights w j are determined by Equation 4

w j = H j ∑ k = 1 m H k (4)

where H k is the cumulative sum of the information measures across all criteria.

3.3 TOPSIS method for ranking alternatives

TOPSIS is employed in this study to handle both benefit and non-benefit criteria to rank locations that are suitable for infrastructure planning for EVBSS.

The following are the steps for ranking the alternatives: first step is to normalise the data value for each criterion. Finding the beneficial and non-beneficial criterion value provides ideal solutions. The most desirable and least desirable solutions are calculated based on the Euclidean distance, which measures geometric closeness. The aggregation and ranking of alternatives are achieved through the preference function, which combines criterion weights with the calculated distance. Based on the outcome list, the alternatives are ranked.

Step 1: The normalised multi-criteria decision matrix is calculated.

A normalised decision matrix is calculated to help in the preference homogeneity of the criteria during the calculation. Where r ij is the normalised matrix (Equation 5).

r ij = X ij ∑ i = 1 n X ij 2 (5)

Step 2: The value of the weighted normalised decision matrix V ij is calculated (Equation 6).

Each element of the r ij was then multiplied by the weights of the corresponding criteria, which had been previously obtained from the CRITIC method.

V ij = r ij × w j (6)

Step 3: The Ideal Solutions are obtained (Equation 7).

The value of A + is obtained from a beneficial criterion; the highest is best, and the lowest is worst. From the non-beneficial criterion, the value of A − is obtained. Where the lowest is best, and the highest is worst. Where A + is a positive ideal solution and A − is a negative ideal solution

A + = { ( max v ij ∣ , ∣ j ∈ J ∣ , ∣ min v ij ∣ , ∣ j ∈ J − ) }

A − = { ( min v ij ∣ , ∣ j ∈ J ∣ , ∣ max v ij ∣ , ∣ j ∈ J − ) } (7)

Step 4: The Euclidean Distance from the ideal solution is calculated (Equations 8, 9).

Mathematically, for an alternative i, the Euclidean distance S i + from the positive ideal solution is calculated as:

S i + = ∑ i n ( V ij − V j + ) 2 (8)

where V j + is the ideal (best) value for the j th criterion and where V j − is the ideal (worst) value for the j th criterion. Similarly, the distance from the negative ideal solution, S i − is calculated (Equation 9).

S i − = ∑ i n ( V ij − V j − ) 2 (9)

Step 5: The relative closeness P i of each alternative to the ideal solution is calculated (Equation 10).

P i = S i − S i + + S i − (10)

Step 6: The final step involves ranking the alternatives.

Based on the relative closeness to the ideal solution, the alternatives are ranked with larger values representing more favourable performance.

3.4 COPRAS method for ranking alternatives

COPRAS method is employed to evaluate the proportional performance and direct contribution of each criterion by handling both positive and negative attributes presented in a study (Zavadskas et al., 2008). The following are the steps for ranking alternatives: first step is to normalise the decision matrix to ensure that all the values remain within a uniform scale, preserving the integrity of the weight allocation.

Step 1: The normalised decision matrix was obtained (Equation 11)

r ij = x ij ∑ i = 1 m x ij (11)

After normalisation, the normalised decision matrix contains values that remove the influence of different units and scales across criteria.

Step 2: The weighted normalised decision matrix is derived (Equation 12).

V ij = r ij × w j (12)

Step 3: The minimising and maximising indices for the alternative are computed (Equation 13).

s i + = ∑ j = 1 k V ij for beneficial criteria.

s i − = ∑ j = k + 1 n V ij for beneficial criteria (13)

The value of s i + is obtained from a sum of beneficial criterion; the greatest value is best, and the lowest is worst. From the sum of non-beneficial criterion value of s i − is obtained where the lowest is best, and the highest is worst.

Step 4: The relative weight ( Q i ) of each alternative is derived according to (Equation 14).

Q i = s + i + min i s i − ∑ j = 1 m s i − s i − ∑ i = 1 m min i s i − s i − (14)

where min i s i − is a minimum of non-beneficial sum.

After finding the Q i, performance index ( U i ) for each alternative is obtained (Equation 15). It reflects the overall utility of each alternative, which considers all criteria and their weights. Alternatives are then ranked based on U_i value. The highest U_i Values indicate a better alternative.

U i = Q i Q max 100 % (15)

Where Q max refers to the maximum relative significance.

4 Calculation and results

4.1 Calculation of criteria weights

A total of 5 criteria were defined, and criterion C1 is considered as beneficial, whereas criteria C2, C3, C4 and C5 are considered as non-beneficial, favouring lower values. After normalising the decision matrix, the Standard Deviation (SDT) for each criterion has been calculated. The SDT of C3 land cost is high (0.401), indicating the strongest differentiating power. Although C4 Proximity to emergency services is an important criterion, its standard deviation is relatively low (0.345), which indicates lower variability among alternatives. The measure of conflict created by the criterion and the critic score is calculated to determine the weights for each criterion, as shown in Table 4.

Table 4. Weights of each criterion.

Although C3 land cost has stronger variability, C1 EV adoption rate achieved the highest CRITIC score due to its negative correlation with other criteria, which indicates that criteria C1 provide more unique and non-redundant information compared to C3.

The criterion weights obtained from the CRITIC method are presented in Table 5. Based on the weights of the criteria with the highest priority, as observed from the results, which can be expressed in a hierarchy as follows: EV adoption rate > land cost > proximity to residential area > safety and security > proximity to public facilities.

Table 5. Criterion ranking.

4.2 TOPSIS method

Once the criteria weights are obtained, the alternative ranks are calculated by applying the criteria weights to the normalised decision matrix using the TOPSIS method. In Table 6, A +and A − are obtained from a beneficial criterion and a non-beneficial criterion. It serves as a benchmark for evaluating alternative performance.

Table 6. Weighted normalised decision matrix with ideal solution.

After obtaining A +and A −, the separation measure S i + and S i −, which represent Euclidean Distance and Performance Score (P i), are calculated and presented in Table 7 to rank the selected alternative location. The S i + and S i − values of L4 indicate closer proximity to the ideal solution and are far from the worst, which attained the highest P i (0.8), ranked first. In the research work, in order to check the reliability of the outcome, the result produced by TOPSIS is compared with the COPRAS method.

Table 7. Rank of selected location.

4.3 Comparative analysis of method - COPRAS

The reliability of the TOPSIS result was evaluated through additional comparison using the COPRAS method. The relative weight for each alternative (S_i+, S_i-), the performance index (U_i) and the rank are calculated as shown in Table 8. The L4 (0.155) has a higher S_i+ value, which indicates better performance in the beneficial criteria, scored higher U_i and ranked first.

Table 8. Rank of selected location.

The COPRAS and TOPSIS rankings of the alternative locations are presented in Table 9, which shows consistency for locations L4, L3 and L1. To evaluate the reliability of the rankings produced by the hybrid MCDM method, Spearman’s rank coefficient is employed as shown below:

Table 9. Spearman’s rank correlation coefficient.

Spearman’s rank r s = 1 − 6 ∑ D 2 n ( n 2 − 1 ) .

where n and D represent the number of alternatives and rank differences, respectively.

The calculated Spearman’s rank coefficient r s is 0.8286. Thus, this reveals a strong correlation between the two methods presented through the scatter chart in Figure 2. Therefore, the comparation of both methods shows that locations L4 and L3 are the most suitable candidate sites for setting up EV BSS inside the CMA.

4.4 GIS-based buffer analysis

The spatial validation was performed using ArcGIS to examine the geographical distribution of the candidate locations within the CMA. The map integrates three components: first, the six identified sites; second, 5-km buffer zones around the locations where TOPSIS and COPRAS produced identical ranks; and third, graduated symbols represent the corresponding EV adoption rates. The purpose of the buffer analysis is to demonstrate spatial representation of ranking consistency, and the graduated symbols provide demand-related spatial indication. The buffer analysis was applied to locations; Tambaram, Poonamallee, and Ambattur exhibited the same ranking. The remaining locations, Tiruvallur, Sriperumbudur and Chengalpattu, were retained as point locations as represented in Figure 3. This combined visualisation facilitates the fact that the MCDM ranks are spatially associated with the existing level of EV adoption across CMA.

4.5 Sensitivity analysis

In this study, a Monte Carlo analysis with 10,000 iterations was performed to assess the robustness of the CRITIC-TOPSIS ranking under uncertainty in criteria weights. The criteria weight was randomly perturbed within ±20% there original values and subsequently normalised during each iteration. The result of the Monte Carlo analysis is presented in Figure 3, through complementary visualisations: a rank frequency heatmap, a mean-rank analysis chart and a best site probability chart. The heatmap illustrates the stability of the ranking position under weight perturbations, while the mean-rank analysis presents rank consistency, and finally, the probability chart enables the consistently preferred alternative, which Tambaram (L4) highlights and confirms the resilience of the model.

The agreement between the deterministic TOPSIS ranking and the Monte Carlo mean ranking was evaluated using Spearman’s rank correlation coefficient, and the calculated r_s = 1.000 indicates perfect agreement (Zou et al., 2003). The analysis indicates that the proposed decision-making model has high stability and applicability. To examine the robustness of BSS ranking against variation in weighting approach, an additional sensitivity analysis was conducted through subjective weights derived from expert judgments. AHP is used to drive subjective weights and integrated with TOPSIS and COPRAS to rank the candidate locations. The results demonstrate a high degree of consistency between subjective and objective weighting approaches as presented in Table 10.

Table 10. Sensitivity analysis: expert-based cross validation.

5 Findings and discussion

The study scrutinised various sites to determine their appropriateness for EV BSS construction in CMA and then subsequently implemented the comparative hybrid MCDM approaches to improve decision-making robustness for location selection. Many studies suggest using real location data to understand real-world scenarios to determine the ideal location (Maghfiroh and Kavirathna, 2023). In the Indian context, Koirala et al. (2022); Krishankumar and Ecer (2024) has examined BSS location selection, although the studies have different infrastructure type, decision frameworks; land cost was identified as a main criterion. For this study, a decision matrix was formed based on the real location data. EV adoption weighted high at 0.320 and serves as the ultimate proxy for potential charging demand. The land cost obtained the second-highest weight (0.264), which highlights its significance over other criteria. EV adoption rate is a novel criterion, which is included in this study. The other social aspects, like proximity to emergency services and residential areas, are also considered, but their weights are lower compared to economic factors (Figure 4).

The integration of EV adoption rate with the MCDM ranking and GIS buffer analysis provides a more comprehensive spatial interpretation of BSS suitability. The TOPSIS and COPRAS methods were employed for ranking the six alternatives, namely Ambattur (L1), Tiruvallur (L2), Poonamalle (L3), Tambaram (L4), Sriperumbudur (L5), and Chengalpattu (L6). The site rankings obtained using the TOPSIS method indicate that Tambaram (L4), with the highest performance value of 73%, is a suitable location for the establishment of BSS with CMA. This area stands out as the best choice due to factors such as EV adoption rate and significant transportation hubs, which collectively create a favourable environment for successful BSS operation. The reliability of the finding is also supported by (Koirala et al., 2022). The combination of MCDM priority and greater EV adoption provides stronger evidence for further EVBSS assessment. The stakeholder thereby minimises the spatial mismatch between where to build and where BSS are needed by preventing stranded assets, thereby reducing range anxiety.

The Poonamalle (L3) ranked second with a performance value of 62%, and a cross-validation using the COPRAS method results in the candidate site Tambaram (L4), which achieved the highest utility degree of 100 and produces minor differences in L2, L5 and L6; the order of the other sites remains stable. Ambattur (L1) obtained the lowest utility value and performance score, which suggests relatively poor performance with respect to the selected evaluation criteria. Spearman’s rank coefficient (r s = 0.8286) confirms strong agreement between COPRAS and TOPSIS results, indicating consistency of the findings. As shown in the heatmap, alternative locations L4, L3, L5 and L1 retained their positions in all runs, while only L2 and L6 exchanged their positions. The mean rank analysis confirms the robustness of the ranking, with L4 and L3 showing the lowest average rank. The probability chart indicates that L4 achieved the first rank in 100% of all iterations. These findings show the stability and robustness of the proposed ranking. From the spatial perspective, the results demonstrate that the candidate locations are distributed across different portions of CMA; this shifts the vehicle-centric model to a mobility-centric sustainable system, which supports SDG 11. From the findings, the robustness of the location prioritisation was confirmed by Monte Carlo simulation and expert-based cross-validation (Garni and Awasthi, 2019; Hafezparast, 2015).

6 Limitations of the study

This study relies on secondary data sources and a restricted set of evaluation criteria, which may not fully capture the influencing factors of electric vehicle charging infrastructure related to battery swapping station (BSS) site selection in rapidly evolving environments. Another limitation includes the geographical focus on the Chennai Metropolitan Area (CMA), where the results and methodology effectiveness may differ when applied to other regions with different socio-economic or technical influencing factors, such as reliability of power, maintenance cost, and traffic conditions. It enhances the precision and applicability of the findings for EV charging infrastructure planning in diverse urban and peri-urban contexts.

Furthermore, the study did not incorporate advanced modelling approaches. Parking availability, while directly relevant to the physical operability of a BSS, was excluded because it requires site-level spatial survey data that could not be systematically compiled for the study region. Renewable energy integration represents an important dimension of sustainability-oriented BSS planning; incorporating it fully requires energy mix modelling at a spatial resolution not attempted here. However, in future studies, using advanced GIS-driven spatial analysis or machine learning could address these gaps.

7 Conclusion and recommendations

Electric vehicle (EV) adoption in the CMA has shown a clear trend of gradual and consistent increase from 2021 to 2025 across multiple localities (Vahan Sewa, 2025). Some areas within the CMA have experienced notably faster growth, reinforcing the overall upward trajectory in EV usage. The study developed a decision-support framework for ranking EVBSS locations within the CMA. The study identified criteria across social and economic dimensions that influence the site selection for establishing EVBSS in the CMA to improve the charging infrastructure. The criteria weightage highlights that the EV adoption rate and land cost dominate the decision-making process. Among the studied locations, Tambaram (L4) ranks highest in suitability, followed by Poonamalle (L3), Tiruvallur (L2), and Chengalpattu (L6). It shows that peripheral nodes play a significant role in EV BSS charging infrastructure planning, which connects the metropolitan centre with regional transportation corridors. It effectively reduces charging congestion inside CMA.

To provide a spatial interpretation of the selected location, ArcGIS-based analysis was incorporated; it provides an additional geographical perspective of EV adoption patterns and the spatial distribution of the selected location. The study’s findings suggest that development in peripheral nodes like Poonamalle (L3), Tiruvallur (L2), and Chengalpattu (L6) helps governments to achieve Sustainable Development Goal (SDG) 11 (Target 11.6 and 11.a), which relates to decentralisation of urban growth, environmental impact reduction and strengthening urban–rural linkages. It reduces the spatial footprint required for long-term parking and plug-in charging. Therefore, in light of the research findings, it is recommended that policymakers and urban planners adopt the proposed MCDM framework to guide data-driven and flexible site selection tailored to local conditions. These analyses demonstrated ranking consistency, particularly the stable first and last positions. Future work may explore integrating grid capacity data, site-level spatial survey data, renewable energy integration and scenario-based demand models as inputs to the weighting and ranking stages.

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: Jingzheng Ren, Hong Kong Polytechnic University, Hong Kong SAR, China

Reviewed by: Burak Oz, Bülent Ecevit University, Türkiye

Sarra Ouerghi, Qatar University, Qatar