ENSEMBLE OF MACHINE LEARNING CLASSIFIERS FOR SCHIZOPHRENIA DETECTION

Abstract

Schizophenia disease is characterized by odd behavior, weird speech and decreased capacity to apprehend reality. The diagnosis of schizophrenia requires a complete and detailed medical examination. Machine learning has also helped computer scientists to classify and diagnose schizophrenia using neuroimaging data. This research implored the use of computer aided diagnosis to classify neuroimaging data of schizophrenia. The dataset of 86 instances which include 40 schizophenia patients, 46 healthy patients and 32 variable. They were obtained from Kaggle MLSF 2014 classification challenge and augmented due to small sized using synthetic minority oversampling technique (SMOTE). This yielded a larger data set of 1806 instances. The augmented data set were classified using machine learning algorithms support vector machine, K-neareast neighbours, logistic regression, Naïve bayes, artificial neural network. 350 instances was used for the training (70%) and 150 instances was used for testing (30%), KNN and SVM correctly classified 162 as Schizophrenia patients and classified 188 as healthy control, Tree correctly classified 159 as schizophrenia, mis-classified 3 as schizophrenia, correctly classified 185 as healthy and mis-classified 3 as healthy control, Logistic Regression correctly classified 139 as schizophrenia, mis-classified 23 as schizophrenia, correctly classified 170 as healthy and mis-classified 18 as healthy control, Naive Bayes correctly classified 139 as schizophrenia, mis- classified 23 as schizophrenia, correctly classified 166 as healthy and mis-classified 22 as healthy control. ANN used 549instances, 60% for training, 20% for testing and 20% for validation got an accuracy of 100%, this makes it the best classification method.