Integrated Machine Learning Algorithms and MCDM Techniques in Optimal Ranking of Battery Electric Vehicles

. The automobile industries across the world of this present age are streamlining the manufacture of battery electric vehicles (BEV) as a step towards creating pollution free environment. BEVs are used as an alternate strategy to alleviate the carbon emission at global level. As environmental conservation is one of the long standing sustainable 1f ?developmental goals it is the need of the hour to make a paradigm shift from fossil fuels to renewable energy sources, at the same time this also gives rise to a decision-making problem on making optimal choice of the electric vehicles. In this paper a decision making problem based on ten alternative BEVs and eleven criteria is considered from the earlier works of Faith Ecer. The new ranking method of multi-criteria decision making MCRAT(Multiple Criteria Ranking by Alternative Trace) is used together with three different criterion weight computing methods of AHP(Analytical Hierarchy Process) ,CRITIC (CRiteria Importance Through Intercriteria Correlation) & MEREC (MEthod based on the Removal Effects of Criteria). The results obtained are compared and validated using random forest machine learning algorithm. This research work conjoins multi-criteria decision making methods and machine learning algorithms to make optimal decisions on Battery electric vehicles and this integrated approach yields optimal ranking results and it will certainly create new rooms in decision-making approaches in coming days.


Introduction
Transportation is quite an inevitable process that employs vehicles of all kinds to displace people and commodity to the required places. Environmental pollution is one of the major impacts of Transportation by means of vehicles based on fossil fuels. The automobile industries across the globe are making stern research on promoting green planet with the introduction of battery electric vehicles comprising different features and accessories. With the vision of attaining the sustainable goal of carbon neutral communities, the production of BEVs is accelerated to reduce and nullify the carbon emission. The sales of electric vehicles at global markets are increasing and it is expected that 30% of the vehicles will turn electric by the end of 2030. Though BEVs are a means of abating pollution, the reluctancy of purchasing electric vehicles still exists amidst buyers. One of the prime reasons is the question on the longevity of the battery used in the electric vehicles. The choice of the battery is highly influenced by several factors such as costs of maintenance, charging time, performance, capacity and many other. Faith Ecer has presented the detailed study on the eleven criteria for choosing the optimal battery and has also described the ranking of ten alternative BEVs using seven MCDM methods with the criterion weights calculated using the method of SECA. Though several decision methods are applied to 10 ×11 decision-making matrix, the criterion weights are computed using only one method. Also the ranking results of Ecer, 2021 [1] obtained using MCDM methods are validated using other MCDM methods of BORDA and COPELAND.
The decision matrix developed by Faith Ecer serves as the motivation for this research work. The methodology developed in this paper is a step to overcome the following research gaps identified in the works of Faith Ecer.


Different methods to calculate criterion weights are not employed to rank the alternatives  Lack of heterogeneity techniques to validate the MCDM ranking results The new initiatives taken in this research work are  Application of three methods to compute criterion weights  Ranking of alternatives using two other different methods of MCRAT and RAPS  Validation of MCDM ranking results using machine learning algorithms The remaining contents are organized into the following subsections as follows, section 2 consists of the literature review of the earlier works, section 3 discusses the methodology, section 4 applies the proposed agglomerated decision making method to the 10 × 11 decision making matrix and section 5 compares the results and concludes the work.

Literature Review
Multi criteria decision making is a process of making optimal decisions based on several criteria. The objective of every MCDM method is to find the optimal ranking of the alternatives using criterion weights. Several methods exist in literature to find the criterion weights and ranking of the alternatives. Researchers  In all the earlier works of MCDM decision on BEVs only very few ranking methods such as TOPSIS, MULTIMOORA, AHP are used. In continuation, Ecer [1], applied seven ranking MCDM methods and only one method of obtaining the criterion weights. The validation of ranking results is done using MCDM methods which are homogeneous in nature. On identification of the research gaps as presented in section 1, a new integrated MCDM method with a machine learning algorithm is developed in section 3 to rank the BEVs. The newly developed integrated MCDM and ML method is a combination of the MCDM methods of criterion weight calculation AHP, CRITIC, MEREC with the ranking MCDM method of MCRAT and the supervised machine learning algorithm of Random Forest Algorithm.
MCRAT is one of the recently developed MCDM methods. This method of ranking the alternatives is based on traces. Urošević et al., [30] has applied this method to rank the alternatives of blasting patterns in mining industry, Gligorić et al., [31] applied to make optimal decisions on coal supplier. In these applications, the usual methods of pair wise comparison are used to obtain the criterion weights. The methods of AHP, CRITIC or MEREC are not used with MCRAT to obtain the criterion weights. This has motivated us to develop an integrated MCDM with the combinations of AHP, CRITIC, MEREC with MCRAT.
AHP is the method developed by Saaty [32], and it is widely applied to obtain the criterion weights. This method comprises three main steps namely hierarchy formation, pairwise comparison and consistency checking. The method of AHP is applied in different fields to determine the criterion weights of the decision making problem.  [44] in truck selection, S. Ghosh and Bhattacharya, [45] in financial performances, Ulas et al., [46] in dealing with banking problems, Fattouch et al., [47] in IoT affairs. These criterion weight computation methods are extensively applied in various other fields in crisp forms, fuzzy and extended fuzzy forms of intuitionistic, Neutrosophic and Plithogenic.
At recent times, the MCDM methods are integrated with machine learning algorithms, especially with the random forest algorithms, one of the most commonly used supervised ML algorithm. In line with it, researchers have used many MCDM methods in combination with RFA, a few recent works are presented as follows, Arabameri et al., [48] in making predictions on gully erosion, Musbah et al., [49] in Energy Management, Srivastava and Eachempati, [50] in employee retention, Q. He et al., [51] in prediction analysis, Pham et al., [52] in land slide vulnerability, Z. He et al., [53] in textile chemical processing, Kadkhodazadeh et al., [54] in climate change , Mustapha et al., [55] in breast cancer screening, Khosravi et al., [56] in flood susceptibility, Z. Wang et al., [57] in chemical engineering , García et al., [58] in credit affairs, Ali et al., [59] in landslide modeling , Pourkhodabakhsh et al., [60] in human resource management, Chowdhury et al., [61] in COVID-19 classification, Baqer et al., [62] in pollutant prediction, Jassim et al., [63] in diagnosis of adult autism, Al-Bawi et al., [64] in gully erosion. In the above mentioned combinations of MCDM and RFA, various decision making methods such as COPRAS, VIKOR, MULTIMOORA are used. To the best of our knowledge the combination of the criterion weight computing methods of AHP, CRITIC, MEREC with ranking method of MCRAT along with Random Forest algorithm do not exist in the literature. This has motivated us to construct an integrated MCDM and ML decision making methods.

Methodology
This section presents the steps involved in finding the criterion weights and ranking procedure of the alternatives. In this the integrated MCDM and ML decision making approach is presented In the above flow chart representation, the decision making matrix is first formulated with alternatives and criteria based on expert's opinion. The criterion weights are determined using any of the methods of CRITIC, AHP or MEREC, then after finding the criterion weights the alternatives are ranked using the method of MCRAT and later validated using Random forest algorithm.

Application of the Integrated Method
In this section, the decision making on optimal alternatives of battery electric vehicles is made based on the following 10×11 decision matrix taken from Faith Ecer. The initial decision making matrix comprises of 10 alternatives and eleven criteria in which six are cost criteria and remaining five are benefit criteria. Let C1,C2……..C11 represents a criteria of BEVs and A1,A2………..A10 denotes Alternatives of batteries. The criteria based on which the ranking of the alternatives is made are presented in Table 1 along with a concise description  The sum of the criterion weights obtained in each of the three methods is equal to one. Fig.1, 2 and 3 presents the distribution of weights to all the eleven criteria.

Fig. 3 Criterion Weights by CRITIC
By using the above criterion weights, the ten alternatives are ranked using the MCDM method of MCRAT as described in section 3. The score values of the ten alternatives are presented in Table 4 and the graphical representation presenting the ranking of the alternatives is given in Fig.4

Fig.4 Score values of the Alternatives
As a means of testing the consistency of the MCDM methods, the machine learning algorithm is used with the criterion weights as given in Table 2. The score values of the alternatives using the method of Random Forest Algorithm is presented in Table 5. The alternatives with score values 0 are more acceptable and the alternatives with maximum score values are highly rejected. The graphical representation of the score values is presented in Fig5. The ranking results of the alternatives based on MCRAT with the combination of three methods of criterion weight computation is presented in Table 6. The maximum score values are given priorities. 10 10 5 The ranking results of the alternatives based on RFA is presented in Table 7. The alternatives are grouped into acceptable and rejectable and in each combination of the criterion weight computation methods the alternatives that fall into two of the groups vary.

Discussion
The ranking results obtained from the applications of different MCDM methods by Faith Ecer is compared with the ranking results of MCRAT AHP, MCRAT CRITIC & MCRAT MEREC. Table 8 presents the comparative analysis of the ranking results. It is found that the ranking results obtained using MCRAT CRITIC is highly in line with almost all the ranking results of Faith Ecer. It is also observed that the ranking results obtained from the method of MCRAT CRITIC is more consistent in comparison with the results of MCRAT AHP & MCRAT MEREC.
Also from the Tables 6 & 7 the ranking results of the alternatives are found to be highly consensus with one another. The results obtained using machine learning algorithms ease the process of finding the acceptable alternatives. On grading the ten alternatives as acceptable and rejectable, the most feasible alternatives are determined. Then on applying the above ranking methods to the acceptable alternatives, the following results are obtained as in Table 9. Also the ranking results obtained using MCDM method and the integrated MCDM & ML methods are compared and it is found that the results of the latter integrated method to be more feasible and the Table 10 vividly present the same. In this case the results are more promising on integrating ML algorithms with MCDM methods. Also the burden of ranking all the ten alternatives is reduced. This will be highly beneficial to the decision makers in handling with many number of alternatives. The initial grouping of the alternatives as acceptable and rejectable using ML algorithms makes the decision process smoother. Then the acceptable alternatives are ranked using the MCDM methods. The optimal ranking of the alternatives thus obtained out of this integrated approach is A4 > A2 > A6 > A1 > A3.

Conclusion
This paper presents an integrated decision method of MCRAT and RFA with three different methods of obtaining criterion weights. The integrated approach simplifies the task of ranking alternatives. The grouping of the alternatives or the classification of alternatives as acceptable and rejectable lessens the difficulties of ordering the alternatives. The method of CRITIC used for determining the criterion weights is highly optimal in comparison with other methods. Also the reduction of alternatives helps in considering only the feasible alternatives and henceforth the ranking of the alternatives is made more easier is the advantage of this integrated approach. This method shall be applied to other decision making problems.