With the rapid development of intelligent connected vehicle technology, vehicles are no longer just means of transport; they have become mobile intelligent terminals that integrate environmental perception, computing and decision-making, and data interaction. Based on big data analysis, this paper studies the methods for extracting driving behavior characteristics, creates a multi-dimensional driving behavior evaluation system, and uses the LSTM-Attention deep learning model to predict driving risks. The model's risk warning effectiveness is verified using real road test data. Driving behavior analysis driven by big data can effectively identify abnormal driving patterns, significantly improve the predictive ability of active safety systems, and provide a theoretical basis and technical support for active safety management of intelligent connected vehicles.