Machine Learning-Based Driving Style Classification Real-World Data Prediction

Driving style remains a critical factor that influences traffic behavior and, hence, risks of accidents even when Advanced Driver Assistance Systems (ADAS) and autonomous driving technologies are increasingly available. In this study, a machine learning-based methodology is proposed for classifying driving into three categories: conservative, normal, and aggressive. Simulation data were used for training the model, although real-world data applications were given for a rigorous evaluation to ensure practical applicability. Several key features, such as acceleration, deceleration, lateral acceleration, throttle input, and brake usage, were analyzed for the classification. The Random Forest model, through comparative analysis, was selected based on better accuracy and generalization. A sliding window technique and statistical feature extraction were used to improve dataset diversity and learning on time-series data. It worked with an 86.7% accuracy rate on real-world data, which sheds light on its robustness for boosting driver behavior analysis as well as personalized driver-assistance technologies.