A Novel Hybrid Algorithm for Automated Schizophrenia Detection using EEG Signals and Deep Learning

The paper proposes a new hybrid approach for EEG-based automated schizophrenia detection that overcomes the limitations of EEG analysis by human specialists and shallow machine learning (ML) methods, which result in low accuracy and manual feature extraction. The proposed method employs the Wavelet Transform (WT) for efficient feature extraction, followed by a Deep Learning (DL) model that combines Convolutional Neural Networks (CNN) for spatial feature extraction and Long Short-Term Memory (LSTM) networks for temporal feature analysis. The synergetic model enhances the accuracy and reliability of diagnosing schizophrenia. The results show that the proposed system's accuracy, sensitivity, and specificity (95.6%, 93.8%, and 96.4%, respectively) clearly surpass the existing one (85.2% and 90.1%). Furthermore, the proposed method reduced false positives (3.2%) and false negatives (4.1%) while achieving the best performance across all parameters and frequency bands, making it an effective tool for clinical early diagnosis applications.