Persona-based stratification and XGBoost modeling for multi-dimensional vehicle sound quality evaluation

Area of Science:

• Automotive Engineering
• Psychoacoustics
• Human-Computer Interaction

Background:

• Electrification and intelligent mobility necessitate advanced vehicle sound quality (SQ) development.
• Traditional NVH methods are insufficient for meeting personalized user expectations.
• Understanding individual differences in SQ perception is key for future automotive design.

Purpose of the Study:

• To investigate the impact of individual differences on multi-dimensional SQ evaluation.
• To develop a robust framework for explaining and reconstructing personalized SQ.
• To provide a mechanistic perspective on population heterogeneity for precise vehicle SQ tuning.

Main Methods:

• Utilized a bipolar 7-point scale for subjective SQ ratings (Comfort, Sportiness).
• Constructed detailed user personas via threshold-based preference splitting.
• Employed Extreme Gradient Boosting (XGBoost) and SHAP (SHapley Additive exPlanations) for nonlinear mapping and feature contribution analysis.

Main Results:

• Developed a robust explanatory reconstruction framework for multi-dimensional SQ.
• Quantified the contributions of psychoacoustic features to Comfort and Sportiness dimensions using SHAP.
• Demonstrated robustness and interpretability of feature contributions across perceptual dimensions.

Conclusions:

• Persona-based stratification effectively reveals population heterogeneity in SQ perception.
• Findings provide a physically interpretable basis for precise vehicle SQ tuning.
• Advanced modeling techniques enhance understanding of subjective sound experiences in automotive contexts.