Prediction of Acoustic Comfort Perception in Electric Vehicles Based on Bayesian Optimized GBDT

To overcome the limitations of machine learning models in predicting acoustic comfort perception in electric vehicles, this study proposes an ensemble learning approach based on Bayesian Optimized Gradient Boosting Decision Tree (BO-GBDT). First, interior noise samples were collected through controlled experiments, and corresponding comfort perception data were obtained via subjective and objective assessments. Subsequently, Bayesian Optimization (BO) was employed to fine-tune the hyperparameters of the GBDT model, aiming to enhance prediction accuracy. A Gaussian process-based surrogate model was constructed to guide the hyperparameter search process. The experimental results demonstrate that the BO-GBDT model exhibits high prediction accuracy, excellent alignment with subjective evaluations, and strong generalization capability, providing valuable support for the optimization of vehicle interior noise and the improvement of sound quality in electric vehicles.