An Interpretable Prediction Framework for Multi-Class Situational Awareness in Conditionally Automated Driving

Hongtao Zheng,Tongtong Zhou,Ting Han,Shuo Li,Cong Yu

DOI: https://doi.org/10.1016/j.aei.2024.102683

IF: 8.8

2024-01-01

Advanced Engineering Informatics

Abstract:Predicting situational awareness (SA) during non-driving-related tasks is critical for improving takeover performance. Although studies have focused on SA prediction based on binary classification, the interpretability of such prediction models is limited, thereby leading to inadequate coverage of SA and suboptimal model efficiency. Therefore, in this study, an interpretable prediction framework based on Extreme Gradient Boosting (XGBoost) and Shapley Additive Explanations (SHAP) was developed for multi-class SA prediction. This model first integrates various information sources, including data regarding SA evaluation, driver eye-tracking (ET), vehicle parameters, and traffic conditions. Second, it quantifies the uncertainty (diversity) produced by SA evaluation using interval-based information granules. SA labels are then generated by considering the bound and number of intervals. Third, the proposed model performs SA classification prediction using XGBoost and investigates the causal relationships among key SA factors by using SHAP. Results revealed that the interaction between ET and vehicle status enhanced SA classification accuracy, and XGBoost outperformed four other models. The findings of the study can be used for the development of safety countermeasures during takeovers.