<p>Vehicle lane change behavior is frequently influenced by drivers’ subjective intentions, leading to randomness and potential traffic conflicts. Especially on highways, hasty lane changes can result in severe crashes due to the high-speed flow of traffic. Aggressive driving poses risks to other road users, both physically and psychologically. To address this, we propose a CNN-GRU model for identifying aggressive lane change behavior on highways. In this study, kinematic data from vehicles was gathered using driving simulation experiments, and lane change events are extracted using a threshold detection method. Driving styles are clustered into three categories: aggressive, normal, and cautious, employing the K-means clustering method. The constructed CNN-GRU model operates in a two-step process. Initially, the CNN structural layer compresses the sample feature map, simplifying network complexity and extracting primary features. Subsequently, the GRU structural layer serves as a recurrent layer to capture temporal characteristics. The fully connected layer then produces recognition results. The model achieved an accuracy of 0.86, with a precision of 0.88, recall of 0.87, and an F1-score of 0.87. Comparing with single CNN, GRU and other machine learning models, the CNN-GRU model significantly improves accuracy in identifying aggressive lane changes. Its combines architecture, leveraging both convolutional and recurrent layers, enhances the temporal feature extraction, resulting in more accurate and reliable detection of aggressive driving patterns. This comprehensive performance is pivotal for accurately identifying aggressive lane change behaviors, ensuring traffic safety, and supporting driving risk assessment and behavior improvement.</p>