Examining Factors Influencing the Acceleration Behavior of Autonomous Vehicles through Explainable AI Analysis

This paper presents a novel approach to understanding the factors influencing autonomous vehicle (AV) acceleration in mixed traffic environments, crucial for the smart transformation of urban mobility systems. The study introduces a pioneering Regularized Stacked Long Short-Term Memory (RS-LSTM) model for predicting AV acceleration. Employing explainable AI techniques, including SHapley Additive exPlanations (SHAP) and Partial Dependence Plots (PDPs), the study interprets factors shaping AV behavior. OpenACC dataset is utilized for model training, testing, and factor exploration. Results reveal that leader acceleration is the most influential factor in determining follower AV acceleration. Additionally, space headway and follower speed exhibit critical thresholds (32 meters and 28 m/s, respectively), beyond which the relationship with AV acceleration predictions undergoes a change. These findings contribute to a deeper understanding of AV behavior in mixed traffic scenarios, with implications for optimizing AV performance and integration in real-world traffic conditions.