With the development of the automotive industry, user experience research of automotive HMI design has gained increasing attention from automotive suppliers. How to discover design issues that have serious impact on user experience in the design stage is an urgent problem that need to solve. This paper proposed a new method for automotive HMI evaluation by combining usability testing data and the XGBoost algorithm. The XGBoost algorithm is utilized to construct a user complaint risk prediction model based on usability testing data. The model achieves an accuracy rate of 85.98% and anAUCvalue of 0.92, demonstrating good predictive performance. Feature importance analysis revealed that ease of use, cumulative driving mileage, task completion time, frequency of use, and task completion status had a greater impact on user experience, whereas path length and function type had less impact. The proposed model in this paper can be further improved in the future by combining expert evaluation methods to achieve more comprehensive and reliable label classification, and more evaluation metrics such as eye tracking data and driving behavior data can be introduced to improve the accuracy and robustness of the evaluation results.
