Detection of schizophrenia using EEG signals: A Machine learning approach

Schizophrenia (SZ) is recognized by the United Nations as a serious mental disorder which is affected by 20 million people worldwide. The symptoms include combinations of hallucinations, delusions, and extremely disordered thinking and behaviour. SZ affects a person in every walks of his life and makes it difficult to move on. Traditionally the diagnosis of SZ is carried by a trained psychiatrist through careful and sharp patient interviews. This process is time consuming and may cause errors. Hence the aim of this work is help clinicians to carry out the diagnosis efficiently. The wavelet transforms which is a signal processing technique is used to analyze the nonlinear and non-stationary EEG signals. In wavelet analysis all the EEG signals has to be decomposed into approximation & detailed coefficients and using these wavelet coefficients four statistical features are calculated. T test class separability criteria is applied on these features and as a result IQR came out to be the best feature. This feature matrix is tested on various well-versed machine learning classifiers out of which SVM showed the best classification rate of 90.14% for SZ and an overall accuracy of 89.58%The results are a clear indication that EEG signals have proved to be a biomarker to discriminate SZ and will be a helpful tool for clinicians in future.