A comparative analysis of regression models and country-level adoption trajectories of electric vehicles

Department of Electrical and Electronics Engineering Technology, Rufus Giwa Polytechnic, Owo, Nigeria

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

With an increase in electric vehicle (EV) adoption at a rapid pace, the need for quality forecasting models to establish different adoption patterns across multiple countries will accommodate for this need moving forward. This work provides analysis between seven regression algorithms for forecasting EV sales and looks to establish the country-level patterns of EV adoption using various clustering techniques. Historical data from the International Energy Agency (IEA) Global EV dataset was used to capture this data at the country level from 2010 through to 2024 across 43 countries worldwide. These historical data were used to perform predictive modelling using the following models; Linear Regression, Ridge Regression, Random Forest, Gradient Boosting, XGBoost, K-Nearest Neighbours (KNN), and Support Vector Regression (SVR), to track statistical performance of each model with respect to predictive modelling through Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and a Coefficient of Determination (R²). In addition, Principal Component Analysis (PCA) and K-Means Clustering were used to identify and track country-level EV adoption trajectories. Of the models used to predict EV sales, Ridge Regression showed the greatest level of forecasting performance with the lowest levels of MAE, RMSE and the least negative R²; while complex ensemble and kernel-based models showed high levels of overfitting as well as poor levels of generalization. The PCA results presented four country-level clusters and highlighted mature markets, high growth emerging economies and low volume adopters of EVs. Overall, the results indicate that using regularised linear models will yield more reliable short-term forecasting than more complex machine learning algorithms.