An Automated MDD Detection System based on Machine Learning Methods in Smart Connected Healthcare

Abstract

Electroencephalography (EEG)-based depression detection in the early stage is a very challenging and important research area in artificial intelligence as it can save the lives of several people. This paper presents EEG-based machine learning models involving 30 healthy subjects and 33 major depressive disorder (MDD) subjects to diagnose MDD. The model with the best performance has been evaluated on the Internet of Medical Things (IoMT) framework for smart healthcare. The main idea behind this study is to recognize features and classifiers which can best discriminate the healthy and depressive subjects. This study has three main steps of analysis: 1) Linear, non-linear, fractal dimension, statistical, time, coherence features have been extracted from EEG signals. Their effects are investigated, and quality features are identified. 2) Three feature selection methods, Principle component analysis (PCA), Neighbourhood component analysis (NBA), and Relief-based algorithm (RBA), are utilized for the selection of most relevant features, and their performance is compared. 3) For discriminating normal and depressed subjects, radial-basis function (RBF) based support vector machine (SVM), K- nearest neighbor (KNN), logistic regression (LR), decision tree (DT), naïve Bayes classification (NBC), bagged tree (BT) and linear discriminant analysis (LDA) classifier are used. This paper concludes that non-linear features with an RBF-SVM classifier achieve the best classification accuracy of 98.90%. The findings in this study are utilized to develop a model to detect depression in remote applications and smart healthcare.