Improving Driving Style in Connected Vehicles via Predicting Road Surface, Traffic, and Driving Style

Abstract

This paper investigates the application of ensemble learning in improving the accuracy and reliability of predictions in connected vehicle systems, focusing on driving style, road surface quality, and traffic conditions. Our study’s central methodology is the voting classifier ensemble method, which integrates predictions from multiple machine learning models to improve overall predictive performance. Specifically, the ensemble method combines insights from random forest, decision tree, and K-nearest neighbors models, leveraging their individual strengths while compensating for their weaknesses. This approach resulted in high accuracy rates of 94.67% for driving style, 99.10% for road surface, and 98.80% for traffic predictions, demonstrating the robustness of the ensemble technique. Additionally, our research emphasizes the importance of model explanation ability, employing the tree interpreter tool to provide detailed insights into how different features influence predictions. This paper proposes a model based on the algorithm GLOSA for sharing data between connected vehicles and the algorithm CTCRA for sending road information to navigation application users. Based on prediction results using ensemble learning and similarity in driving styles, road surface conditions, and traffic conditions, an ensemble learning approach is used. This not only contributes to the predictions’ transparency and trustworthiness but also highlights the practical implications of ensemble learning in improving real-time decision-making and vehicle safety in intelligent transportation systems. The findings underscore the significant potential of advanced ensemble methods for addressing complex challenges in vehicular data analysis.