Comparing the Performance of Classification Algorithms for Predicting Electric Vehicle Adoption

Abstract

In this study, the Electric Vehicle (EV) purchase decisions of European consumers are predicted using supervised machine learning (ML), specifically classification. Following the replacement (imputing) of missing data values through predicted values and continuizing of all predictor features, the predictor features are ranked according to the Information Gain Ratio and the Gini coefficient. The results suggest that suiting daily driving needs (Q17), belief that society must reward electric cars instead of petrol and diesel cars (Q14), and opinion change regarding electric cars during the past year (Q21) ranked the highest with respect to the Gini coefficient metric. The same predictor features rank the highest with respect to the Information Gain Ratio metric, yet in a different rank (Q17, Q21, and Q14). For predictive analytics, a multitude of classification algorithms are applied to predict the decision of EV purchase, and the performance of the applied algorithms is compared. The results suggest that gradient boosting performed best in predicting EV adoption decisions, followed by the logistic regression and random forest algorithms.