NeuroRAG: An Explainable EEG-Based Stress Detection System Using Retrieval-Augmented Generation

Contemporary EEG-based stress detection systems produce binary outputs without clinical explanation. This paper presents NeuroRAG, a system combining EEG biomarker extraction (TAR, FAA, EI), clinical narrative generation, and Retrieval-Augmented Generation (RAG) for explainable stress assessment. Evaluated on 112 EEG recordings across 5 mental state conditions (SCWT, CMPS, TMCT, HORROR, MUSIC), the system confirms SCWT produces the highest cognitive stress (mean=0.502, 33.3% HIGH_STRESS) while Horror video stimulation yields the lowest (mean=0.302, 9.1% HIGH_STRESS), correctly distinguishing prefrontal cognitive stress from emotional arousal. Retrieval accuracy is 24.1% (27/112) with neurophysiologically coherent confusion patterns. To the best of the author's knowledge, this is the first system combining EEG biomarker embeddings, clinical narrative generation, and RAG-based LLM reasoning for explainable stress detection.