Kansei analysis for interior design of new energy vehicles: combining rough set theory, eye tracking, and improved gated recurrent unit

To address inaccurate translations of Kansei needs in New Energy Vehicle (NEV) cabin experiences, this study proposes a universal NEV interior design decision-making framework mapping design features to fuzzy Kansei imagery. Core semantics are extracted via factor analysis. We introduce an Attention-Aware Rough Set Feature Selection Method (ARFS) utilising Rough Set Theory for design feature reduction, integrating Eye-Tracking to capture visual search paths. The reduced design features are reorganised based on attention sequences to align with visual cognitive patterns. Next, a TCN-BiGRU-Attention deep learning model is used to fits NEV interior design features with Kansei data, where Temporal Convolutional Networks extract local non-linear couplings, Bidirectional Gated Recurrent Units capture long-range contextual dependencies, and an Attention mechanism dynamically assigns feature weights. 5-fold cross-validation and ablation experiments demonstrate that ARFS significantly improves prediction compared to single modes, and demonstrating its transferability across typical NEV interior components. Furthermore, to overcome single-solution limitations and mitigate prediction bias of traditional Kansei engineering, we establish an uncertainty-aware recommendation mechanism. By using 5-fold cross-validation to quantify epistemic uncertainty, the system provides designers with a candidate set of top-ranked and statistically significant options, successfully balancing design diversity with the fault tolerance needed for engineering decisions.