Classification of Schizophrenia Patients on the Basis of Latent Components of ERPs using a Machine Learning System

Schizophrenia is an endogenous mental disorder leading to disability. Timely treatment is crucial for mitigating long-term negative effects. In this study, event-related potentials (ERPs) obtained from EEG data of patients and healthy individuals performing a modified visual Go/NoGo task were analyzed. We used the method of blind source separation (BSS) based on second-order statistics to decompose ERPs into functionally distinct components. Various features were extracted from these latent components within specific time windows for use in machine learning models. The Support Vector Machine (SVM) model achieved sensitivity and specificity of 96.7% and 97.7%, respectively. Using SHAP method, the most significant features were identified to understand the model's decision-making factors. These factors were validated by experts in physiology, aligning with their expectations.