A Machine Learning Framework for Automatic Diagnosis of Schizophrenia Using EEG Signals

Schizophrenia (ScZ) is a chronic brain disorder that affects speech, mood, behaviour, cognitive ability, etc. The people suffering from this disease often misinterpret reality, lose concentration in daily activities, and always keep themselves in a state of anxiety, depression, and confusion. Manual inspection of the patient by a well-trained psychiatrist is quite subjective, and time-consuming task. The need for long-term medicine in the treatment of mental illness places an undue financial burden on families. Therefore, early intervention and proper care is crucial in planning the treatment for patients. In this work, an automated identification model has been developed to differentiate normal healthy adolescents and Schizophrenia symptomatic adolescents from the EEG signals. The proposed methodology is demonstrated on the openly accessible EEG dataset which includes 16-channel EEG signals from 84 adolescents (45 SZ and 39 healthy) released by Lomonosov Moscow State University. Initially, signal preprocessing is done, and then 24 features (statistical, non-linear, and other EEG features) are extracted from EEG data. Among them, only 8 features are designated as the crucial feature by applying Kruskal Wallis (KW) test. Popular classifiers are used to assess the effectiveness of the selected features. The ensemble bagged tree classifier outperforms the existing methods on the considered EEG dataset with a classification accuracy of 92.3% with 10-fold data division protocol. Hence, this proposed method can effectively distinguish ScZ patients from normal healthy participants and potentially become a tool for psychiatrists to assist in efficiently diagnosing schizophrenia.