EEG-based multi-level stress classification with and without smoothing filter

Recently, multi-level stress assessment has become an active research subject. In this context, researchers typically develop models based on machine learning classifiers and features extracted from biosignals like electrocardiogram (ECG) or electroencephalogram (EEG). For that purpose, EEG power spectral density (PSD) is a recurrent feature owing to its high responsiveness remarkable performance. However, PSD usually smoothed cope with bursty nature, what may cause data leakage hence call into question classification study, our aim was twofold: first, examine the effect of EEG-PSD smoothing in three-level classification, second, evaluate practical viability two-level detector without smoothing. To end, we conducted participants through stress-relax session while recording their EEG. Then, estimated used reported by as labels for classification. Initially, developed classifier examined We found performance directly proportional intensity (F1-score 0.61–0.94), also when not applied features, insufficient applicability (AUC < 0.7). link behavior train-test contamination due Subsequently, attempted case, met criteria = 0.76). This suggest enhancement caused produced smoothing, hence, render realistic each epoch should be processed individually.