Explainable Prediction Study of Design Creativity Levels for the Embodied Interaction Scenario

To enhance the accuracy of predicting designers’ creativity level and gain a deeper understanding of the key neurophysiological features influencing creative thinking, a Bidirectional Long Short-Term Memory (Bi-LSTM) model combined with SHapley Additive exPlanations (SHAP) was developed to predict the creativity level during the design process. The multimodal neurophysiological data was collected in this experiment, including electroencephalogram (EEG), electrodermal activity (EDA), heart rate (HR), and skin temperature (ST) time-series signals, from 34 designers during their creating process in an embodied interaction scenario. The creativity of the design solutions was evaluated using the Consensus Assessment Technique (CAT). Through feature extraction and selection, a creativity level prediction model based on Bi-LSTM was constructed. Performance evaluation results demonstrate that the Bi-LSTM model outperformed five other machine learning algorithms in predicting creativity level, achieving an accuracy of 0.863, a precision of 0.868, a recall of 0.863, and an F₁-score of 0.862. Explain ability analysis based on SHAP values further revealed the contribution patterns of neurophysiological features to creativity level prediction. The results indicate significant differences in the contribution of neurophysiological features in the embodied interaction scenario, with the same features potentially contributing in opposite directions across different individuals.