Intelligent electroencephalogram feature engineering for rapid mental health diagnosis

Schizophrenia is one of the serious disorders and, if left untreated, can result in a range of problems with cognition, behavior, and emotions that affect every area of life. Diagnosis based on behavioral and clinical investigations remains difficult with schizophrenia symptoms which are complex and heterogenic. Early detection of schizophrenia is essential for the timely treatment leading to betterment of the life of patients. In this study based on machine learning algorithms, we have identified the relevant set of features from the electroencephalogram (EEG) signal to improve the classification accuracy of patients with schizophrenia and healthy controls. Combinations of these identified relevant features have been used to diagnose schizophrenia.Furthermore, we validated this same feature set as the high performing feature subset on an independent dataset, confirming its robustness and generalizability. The results show that the selected features from the EEG signal achieve the highest accuracy of 94.7% and 96.4% for Logistic Regression (LR) and Support Vector Machines (SVM) respectively with reduced data. Reduction in training data with this feature selection enhances the performance of edge devices that are optimized for applications such as brain computer interfaces, neurological disorder detection, cognitive state monitoring, and neurofeedback training.