EEG-based Schizophrenia Detection Using Spectral, Entropy, and Graph Connectivity Features with Machine Learning

Abstract Schizophrenia is a serious mental disorder that changes the way people think, perceive, and manage daily life. Getting the diagnosis right is critical for proper treatment, but in practice it is often difficult. Current evaluations depend mostly on a clinician’s judgment, and the overlap of symptoms with bipolar disorder or major depression makes the task even harder. EEG offers a safe and noninvasive way to study brain activity, yet no single EEG feature has been reliable enough to stand on its own. This makes it important to look at integrative approaches that bring together different aspects of brain dynamics. In this study, we analyzed EEG features to distinguish patients with schizophrenia from healthy controls. Spectral power was measured across δ, θ, α, β, and γ bands. Temporal irregularity was quantified with Multiscale Permutation Entropy (MPE), which to our knowledge represents the first application of MPE to EEG in schizophrenia. Functional connectivity was estimated with the weighted Phase Lag Index in θ, α, and β bands, followed by extraction of graph measures including global efficiency, clustering coefficient, characteristic path length, and mean strength. These features were used to train Random Forest, Multi-Layer Perceptron, and Support Vector Machine classifiers. Among the models, Random Forest achieved the most reliable performance, reaching 99.7% accuracy under stratified 5-fold validation and 99.6% under leave-one-subject-out validation. Feature analysis showed that connectivity in θ and α bands contributed most strongly to classification. Topographic maps of θ, α, and β activity also revealed regional group differences. Overall, the results suggest that combining spectral, entropy, and connectivity measures offers a promising framework for EEG-based detection of schizophrenia. Nevertheless, these findings are preliminary given the limited sample size (N=28), and replication in larger and more diverse cohorts is required before clinical translation. Competing Interest Statement The authors have declared no competing interest. Footnotes nazila.ahmadii{at}ut.ac.ir ksetareh{at}ut.ac.ir