EEG-Based Schizophrenia Classification with Machine Learning: Comparative Analysis of CNN, Random Forest, and SVM

Abstract

Schizophrenia is a serious psychiatric condition that lacks reliable indicators, complicating objective diagnosis. Electroencephalography (EEG) offers a noninvasive insight into cerebral function and identifies distinctive irregularities in schizophrenia, indicating its potential for automated diagnosis. This study develops machine learning algorithms to autonomously classify schizophrenia patients utilizing EEG data. We propose and assess three categorization methodologies: a deep Convolutional Neural Network (CNN) and two conventional methods, Random Forest (RF) and Support Vector Machine (SVM). We delineate thorough EEG preprocessing, encompassing artifact elimination and feature extraction (e.g., spectral band power and connection metrics). We employed a publicly accessible EEG dataset comprising schizophrenia patients and healthy controls to train and evaluate these models through cross-validation. The CNN attained the highest accuracy (about 90%), surpassing the RF (around 85%) and SVM (between 80 and 82%). We examine the relative performance, emphasizing the CNN's capacity to identify intricate patterns and the interpretability and efficiency benefits of classical methods. Our findings illustrate the potential of EEG-based machine learning in facilitating schizophrenia diagnosis, while emphasizing the necessity for more validation and advancement for clinical implementation.