Diagnosis of Schizophrenia Using Electroencephalography Signal Analysis an‎d Artificial Intelligence Algorithms

10/23/2025 12:00:00 AM

This thesis investigates an‎d analyzes EEG signals for the diagnosis of schizophrenia. The main objective of this research was to design an EEG-based diagnostic system to identify individuals with schizophrenia in two different age groups: adults an‎d adolescents. In this study, a series of signal preprocessing techniques, as well as time-domain, frequency-domain, an‎d time–frequency feature extraction methods, were applied. Dimensionality reduction algorithms were also employed to improve the accuracy of machine learning models. Furthermore, various machine learning models, including MLP, SVM, k-NN, DT, Bagging, Gradient Boosting, an‎d Voting, were trained to analyze the data, an‎d their performances were eva‎luated using multiple metrics. The results demonstrated that combining features extracted from different domains—particularly time an‎d time–frequency—enhanced the performance of the models in schizophrenia detection, especially in ensemble models such as Voting an‎d Gradient Boosting. Moreover, the application of dimensionality reduction techniques like PCA significantly improved model accuracy. In the adult dataset, the Gradient Boosting model achieved an accuracy of 0.98, while the Voting model reached 0.99. In the adolescent dataset, both the Voting an‎d Gradient Boosting models achieved an accuracy of 0.94 when PCA-based dimensionality reduction was applied. Considering that the duration of EEG recordings for each adult subject exceeded 15 minutes, the signals were segmented into 5-, 10-, 30-, an‎d 60-second windows. Based on ROC_AUC comparisons, the 30-second window was identified as the optimal time segment length for the adult dataset, providing the best balance between classification accuracy an‎d model stability. This thesis can serve as an initial step toward the development of EEG-based clinical diagnostic systems for use in hospitals an‎d medical centers.

سيگنال EEG , اسكيزوفرني , يادگيري ماشين

EEG signal , schizophrenia , machine learning

Hossein Mehralipoorfard

Dr Ali Sadr