Advanced EEG Signal Preprocessing and Decomposition: A Multi-Stage Pipeline for Neurological Disorder Detection

Researchers and clinicians widely use electroencephalography signals for non-invasive assessment of neurological function; however, artifacts, broadband noise, non-stationarity, and high inter-subject variability affect their clinical interpretation. This study proposes a multi-stage EEG preprocessing and decomposition framework for detecting neurological disorders. We evaluate the framework on three public EEG datasets: CHB-MIT, Sleep-EDF Expanded, and Bonn EEG. Experimental results show an average SNR improvement of 18.5 dB, artifact rejection above 96%, and four-class classification accuracy of 96.2%. The method also achieves 95.7% sensitivity, 96.8% specificity, and an AUC of 0.987. These results suggest that systematic preprocessing and decomposition can substantially improve the reliability of automated EEG-based detection of neurological disorders.