Jiawen Chen, Xuesong Wang*, Mengjiao Wu, Xin Yi, Xiaowei Tang, Andrew Morris
Abstract: Optimizing user-centered alerting systems is essential as automotive technology continues to evolve. However, previous studies have not fully clarified how individual driver characteristics affect the perception and response to warning signals. Consequently, this study employed Random Forest Regression and SHAP analysis to identify significant features and their contribution to predictions. Results showed that lane position, fixation times, and subjective urgency score were strong predictors of brake reaction time. In contrast, subjective pleasantness, driver gender, and subjective urgency score played a major role in perceived subjective warning effectiveness. Lane departure directly influences braking response, while driver characteristics impact subjective warning effectiveness. These findings provide insight into feature selection and model generalizability. They also help to identify the factors that improve the effectiveness of in-vehicle warnings and support safer driving behavior.
引用：Jiawen Chen, Xuesong Wang*, Mengjiao Wu, Xin Yi, Xiaowei Tang, Andrew Morris. Optimizing in-vehicle warning sounds: core feature insights with machine learning models. Transportation Research Part F: Traffic Psychology and Behaviour, Volume 115, November 2025, 103332.