Abstract: Aiming at the problem of sound quality feature modeling, a sound quality prediction method combining time domain analysis and machine learning is proposed. Firstly, the sample entropy is introduced, and the time-domain features of noise signals are extracted by combining the grey wolf optimization (GWO) and variable mode decomposition (VMD) to construct an objective parameter of sound quality. Secondly, in order to improve the vehicle interior sound quality mapping accuracy based on the extreme gradient boosting (XGBoost) algorithm, the adaptive weight (AW) and adaptive factor (AF) for particle swarm optimization (PSO) algorithm are improved, and sound quality modeling methods based on AW-PSO-XGBoost, AF-PSO-XGBoost and AWF-PSO-XGBoost are proposed. Ultimately, the training and testing results of 64 sets of electric bus interior sound quality data indicate that the determined AWF-PSO-XGBoost model has the best acoustic comfort prediction accuracy and fitting effect, with average relative error and consistency coefficient being 3.27% and 0.9889, respectively.