AI-POWERED STRESS MONITORING USING MULTIMODAL IN EDUCATION

Abstract

TriS-Mind integrates NeuroClean, DeepSenseX, and MindFusionNet into one system to create an accurate and reliable multimodal mental state detection system through its pipeline. To perform adaptive preprocessing on NeuroClean’s data, it uses expressive variation preserving normalization, emotional anchor based timestamp alignment, modal graph reconstruction for missing data and noise aware filtering. Additionally, DeepSenseX extracts deep hierarchical representations from audio and image signals (with CNN-Transformer models) as well as physiological data (with Temporal ConvNets) and text data (with Emotional-BERT). After this embedding, all modalities are mapped into the same latent mental state through reliability aware harmonisation (the aggregation of multiple modality embeddings). Finally, MindFusion Net makes the final stress prediction by exploitation of temporal emotional memory; using hierarchical cross attention and dual head output for continuous intensity scores and categories stress level.