Integrating Neurocognitive Signals and Machine Learning for Trust Prediction in Human–AI Collaborative Design

Human trust in artificial intelligence (AI) is critical for decision-making in engineering and design, where AI helps engineers make optimal choices and improve project outcomes. This study combines neuroimaging [i.e., functional near-infrared spectroscopy (fNIRS)] data with interactive experience feedback from human–AI collaborative truss structure design to enhance machine learning models predicting trust behaviors in human–AI interaction. By investigating neurocognitive signals in the prefrontal cortex (PFC) and analyzing participant interactions with AI suggestions, the study evaluated six models: random forest (RF), decision tree, support vector machines (SVM), gradient boosting, quadratic discriminant analysis (QDA), and linear discriminant analysis (LDA). Combining fNIRS data with human–AI interactive experience significantly increases trust behavior prediction accuracy (i.e., accept or reject AI suggestions). Random Forest performed best, indicating 87% accuracy. These findings show the potential of integrating neurocognitive data, interactive experience, and machine learning to advance trust prediction, enhancing human–AI collaboration and decision-making in engineering applications.