Single-channel wearable EEG for ADL recognition in Healthy and Epileptic cohorts using MODWT-MRA and machine learning

Wearable single-channel EEG is increasingly used to monitor epileptic patients outside the clinic, but the physical activity of daily living (ADL) continuously shapes the recorded signal. This activity-coupled variability is typically discarded as noise, even though it holds recoverable contextual information. Until now, EEG-based human activity recognition (HAR) has been studied almost exclusively in healthy participants under controlled lab conditions, leaving ADL recognition in clinical epileptic populations unaddressed. This study introduces a single-channel frontal EEG framework for ADL recognition, evaluated jointly on healthy, epileptic, and combined cohorts. As ambulatory EEG is multicomponent and nonstationary, a five-level maximal overlap discrete wavelet transform-multiresolution analysis (MODWT-MRA) with the Coiflet-3 wavelet decomposes each epoch into five frequency-resolved detail components, from which a compact 55-dimensional handcrafted descriptor is extracted and classified using an ensemble learner. Two recognition granularities are considered: a three-class intensity scheme (stationary, light ambulatory, intense ambulatory) and a fine-grained eight-class individual-ADL scheme, tested across multiple temporal window lengths. Using class-balanced, recording-level holdout evaluation, the framework achieves three-state activity recognition accuracies of 79.7%, 89%, and 85.2% for the Healthy, Epileptic, and Combined cohorts, respectively, with the 5-second window consistently delivering strong performance across all three groups. The method is further tested on the eight harmonized individual ADLs, offering additional insight at the activity-specific level. To our knowledge, this is the first framework to jointly perform EEG-based ADL recognition in both healthy and clinical epileptic cohorts, extending single-channel activity profiling from controlled healthy settings to activity-aware wearable monitoring in clinical populations.