Probabilistic Prediction of Longitudinal Trajectory Considering Driving Heterogeneity With Interpretability

To promise a high degree of safety in complex mixed-traffic scenarios alongside human-driven vehicles, accurately predicting the maneuvers of surrounding vehicles and their future positions is a critical task and attracts much attention. However, most existing studies focus on reasoning about positional information based on objective historical trajectories without fully considering the heterogeneity of driving behaviors. Besides, previous works have focused more on improving models’ accuracy than investigating their interpretability to explore the extent to which a cause and effect can be observed within a system. Therefore, this article proposes a personalized trajectory prediction framework that integrates driving behavior feature representation to account for driver heterogeneity. Specifically, based on a certain length of historical trajectory data, the situation-specific driving preferences of each driver are identified, where key driving behavior feature vectors are extracted to characterize heterogeneity in driving behavior among different drivers. The proposed LSTMMD-DBV (long short-term memory and mixture density networks with driving behavior vectors) framework integrates driving behavior feature representations into a long short-term memory encoder–decoder network to investigate its feasibility and validate its effectiveness in enhancing predictive model performance. Finally, the Shapley Additive Explanations method interprets the trained model for predictions. After experimental analysis, the results indicate that the proposed model can generate probabilistic future trajectories with remarkably improved predictions compared to existing benchmark models. Moreover, the results confirm that the additional input of driving behavior feature vectors representing the heterogeneity of driving behavior could provide more information and, thus, contribute to improving prediction accuracy.