Data-driven machine learning techniques for fuel economy prediction in sustainable transportation systems
Sohaib Zahid, Umar Jamil
Abstract
Sustainable transportation aims to reduce greenhouse gas emissions and improve air quality. While developed countries focus on transitioning to electric vehicles, undeveloped and some developing countries face challenges due to energy crises and high costs, making immediate adoption difficult. In response to the growing demand for vehicles, automotive industries are encountering diverse challenges, including high initial costs due to the integration of intelligent technologies for enhanced vehicle performance. Concurrently, consumers prioritize vehicles with improved fuel economy, aiming to minimize fuel expenses and mitigate environmental impacts like air pollution. The fuel economy of vehicles depends on different features such as their vehicle class, engine size, cylinders, fuel type and city fuel consumption. In this research work, various Machine Learning (ML) techniques such as Linear Regression (LR), Random Forest Regression (RFR), and Support Vector Regression (SVR) are used to predict the fuel economy of vehicles based on the factors mentioned above. After a comparative study, the RFR demonstrated superior performance compared to other machine learning models, such as LR and SVR, using most of the input features. Specifically, with city fuel consumption (L/(100 km)) as the input, RFR achieved a Mean Squared Error (MSE) of 0.839,4, a Mean Absolute Error (MAE) of 0.66, and an R-Squared ( R 2 ) score of 0.984,3 on the 2000–2022 dataset. In comparison, LR resulted in an MSE of 7.375,4, an MAE of 1.754,9, and an R 2 score of 0.856,7, while SVR yielded an MSE of 0.976,1, an MAE of 0.69, and an R 2 score of 0.981,9. On the validated 2023–2024 dataset, RFR maintained superior performance with an MSE of 0.848,6, an MAE of 0.66, and an R 2 score of 0.984,9. In contrast, LR achieved an MSE of 10.504,5, an MAE of 1.950,7, and an R 2 score of 0.827,3, whereas SVR obtained an MSE of 1.104,7, an MAE of 0.75, and an R 2 score of 0.980,9. • Conducted comprehensive data analysis and correlation analysis to explore key factors influencing fuel economy. • Developed and evaluated three machine learning models (RFR, LR, and SVR) for fuel economy prediction. • Performed a comparative analysis of the models' performance, which demonstrated that the RFR model significantly outperforms the LR and SVR models in predicting fuel economy.