Mathematics and Computer Science: Research Updates Vol. 12

Schizophrenia is a severe psychiatric disorder that affects cognition, perception, emotion and behaviour, and timely diagnosis is important for planning appropriate clinical management. Conventional diagnostic approaches are mainly based on clinical interviews and behavioural assessment, which may be influenced by subjectivity and inter-clinician variation. This study proposes an automated EEG-based schizophrenia detection framework that integrates EEGNet, an attention mechanism and Bidirectional Long Short-Term Memory (BiLSTM) networks. EEG recordings from healthy individuals and patients diagnosed with schizophrenia were preprocessed using filtering, Independent Component Analysis-based artefact removal, epoch segmentation and normalisation. Subject-wise data splitting was used to reduce the risk of data leakage during model evaluation. EEGNet was applied to extract spatial and spectral features from multichannel EEG signals, while the attention mechanism assigned greater weight to informative EEG channels and temporal regions. BiLSTM layers were then used to model bidirectional temporal dependencies in the extracted EEG feature sequences. The proposed framework achieved a reported classification accuracy of 99.41% within the described experimental setting. The findings indicate that combining EEGNet-based feature extraction, attention-based feature weighting and BiLSTM-based sequence learning may improve automated schizophrenia classification from EEG signals. However, further validation using independent datasets and detailed reporting of experimental conditions are required before clinical application can be considered.