A Gaussian Filtering Approach for Accurate Detection of Schizophrenia

Schizophrenia (SZ) is a severe disorder affecting the brain functioning. SZ patients tend to exhibit an irrational behaviour. A timely diagnosis becomes essential in controlling the disease and providing necessary treatment to SZ patients. Many recent works have explored low-cost SZ detection using electroencephalogram (EEG). In this work, we develop an accurate and efficient system to detect SZ by dividing EEG signals into suitable sub-band components. The signal decomposition is performed using a set of Gaussian filters, and statistical features are extracted out of these signal components. The significant features are selected by deploying the Kruskal-Wallis test. Classification is performed using support vector machine (SVM) and k-nearest neighbour (kNN) classifiers. Different kernel functions are considered for SVM, while various distance metrics are explored for kNN classifier. The classification results are computed on a popular public dataset and compared with existing state-of-the-art techniques. The ability of the model to generalize for new data is evaluated by employing subject-wise cross-validation strategy, yielding an average accuracy of 90.10% for the selected kNN model with City block distance metric. The findings of this work may be utilized by healthcare professionals for efficient SZ detection. The proposed method is easy to implement and has the potential for widespread practical deployment to detect SZ from EEG signals.