Design of a driving behavior recognition system for road transport vehicles based on deep temporal feature learning and density-based clustering

To address the challenges of insufficient data utilization and limited model accuracy in road transport safety supervision, this study develops a driving behavior recognition system based on vehicle trajectory data, integrating deep feature extraction with density-based clustering analysis. The system adopts a hierarchical architecture comprising data, model, web service, and application layers, realizing the complete process from data acquisition to risk identification and visualization. Methodologically, an LSTM model is employed to extract temporal features from trajectory data, effectively capturing the dynamic patterns of driving behavior. The DBSCAN algorithm is then applied to cluster the extracted feature vectors. Compared with K-Means, GMM, and hierarchical clustering, DBSCAN achieves the best performance with a Silhouette Coefficient of 0.47 and a Davies–Bouldin Index of 0.82, revealing three typical driving styles: aggressive, moderate, and conservative, which correspond to high-risk, medium-risk, and low-risk driving modes, respectively. The results demonstrate that the proposed system exhibits good scalability and provides robust technical support for safety monitoring and driving risk assessment in road transport enterprises.