An Interpretable Decision Modeling and Machine Learning-Driven Method for Bio-Inspired Styling Generation of New Energy Vehicles
- 1College of Design and Innovation, Tongji University, Shanghai 200092, China
Abstract
New energy vehicles (NEVs) are central to sustainable mobility. Existing Kansei Engineering (KE)-driven bio-inspired styling studies often predict optimal Kansei-feature combinations and then rely on subjective biological association, leaving the translation from design features to styling expression insufficiently explained. This study proposes an interpretable decision-modeling and machine-learning framework for NEV side-profile design. User Kansei evaluations and satisfaction data were collected. At the Kansei-need level, grey relational analysis (GRA) ranked associations between Kansei descriptors and satisfaction, while extreme gradient boosting with Shapley additive explanations (XGBoost-SHAP) explained contributions to satisfaction prediction. At the styling-feature level, sparrow search algorithm-optimized support vector regression (SSA-SVR) mapped encoded side-profile features to priority Kansei responses; leave-one-covariate-out (LOCO) analysis identified influential styling features; and accumulated local effects with bootstrap resampling (Bootstrap-ALE) revealed nonlinear predictive effects. These explanations were translated into feature-level bio-inspired rules and three concepts derived from cheetah, dolphin, and frog imagery. Results showed that the three main Kansei needs were Agility, Simplicity, and Freshness, corresponding to the key design features of wheel rim, body shape, and window outline; headlight, side mirror, and taillight; and body shape, window outline, and side mirror, respectively. User evaluation indicated that the three bio-inspired concepts aligned with their respective target attributes, and all three concepts received satisfaction ratings significantly above the scale midpoint. The framework advances KE-driven bio-inspired design toward a traceable, interpretable feature-level translation mechanism for differentiated NEV styling innovation.
Related
- Are Good Generators Good Decision-Makers? Policy Learning for General Interventions via Retargeted Counterfactual Generation
- Interpretable and Validatable Uplift Modeling
- A Quantum-Inspired Dequantization Method for Diagonally Weighted Matrix Functions: Application to Learning with Optimized Random Features
- A lightweight blockchain-inspired hybrid intrusion detection system with ensemble learning for tamper-proof auditing
- Multichoice Modeling: Exploring Decision Configuration Space under Uncertainty
BibTeX
@article{chen2026interpretable,
title = {{An Interpretable Decision Modeling and Machine Learning-Driven Method for Bio-Inspired Styling Generation of New Energy Vehicles}},
author = {Zimo Chen and Jingwen Tian and Duan Wu},
journal = {Mathematics},
year = {2026},
month = nov,
doi = {10.3390/math14214079},
url = {https://www.mdpi.com/2227-7390/14/21/4079},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - An Interpretable Decision Modeling and Machine Learning-Driven Method for Bio-Inspired Styling Generation of New Energy Vehicles
AU - Zimo Chen
AU - Jingwen Tian
AU - Duan Wu
PY - 2026
DA - 2026/11/06
JO - Mathematics
DO - 10.3390/math14214079
UR - https://www.mdpi.com/2227-7390/14/21/4079
AB - New energy vehicles (NEVs) are central to sustainable mobility. Existing Kansei Engineering (KE)-driven bio-inspired styling studies often predict optimal Kansei-feature combinations and then rely on subjective biological association, leaving the translation from design features to styling expression insufficiently explained. This study proposes an interpretable decision-modeling and machine-learning framework for NEV side-profile design. User Kansei evaluations and satisfaction data were collected. At the Kansei-need level, grey relational analysis (GRA) ranked associations between Kansei descriptors and satisfaction, while extreme gradient boosting with Shapley additive explanations (XGBoost-SHAP) explained contributions to satisfaction prediction. At the styling-feature level, sparrow search algorithm-optimized support vector regression (SSA-SVR) mapped encoded side-profile features to priority Kansei responses; leave-one-covariate-out (LOCO) analysis identified influential styling features; and accumulated local effects with bootstrap resampling (Bootstrap-ALE) revealed nonlinear predictive effects. These explanations were translated into feature-level bio-inspired rules and three concepts derived from cheetah, dolphin, and frog imagery. Results showed that the three main Kansei needs were Agility, Simplicity, and Freshness, corresponding to the key design features of wheel rim, body shape, and window outline; headlight, side mirror, and taillight; and body shape, window outline, and side mirror, respectively. User evaluation indicated that the three bio-inspired concepts aligned with their respective target attributes, and all three concepts received satisfaction ratings significantly above the scale midpoint. The framework advances KE-driven bio-inspired design toward a traceable, interpretable feature-level translation mechanism for differentiated NEV styling innovation.
SN - 2227-7390
ER -