Major depressive disorder assessment via enhanced k-nearest neighbor method and EEG signals

The aim of this paper is to introduce a novel method using short-term EEG signals to separate depressed patients from healthy controls. Five common frequency bands (delta, theta, alpha, beta and gamma) were extracted from the signals as linear features, as well as, wavelet packet decomposition to break down signals into certain frequency bands. Afterwards, two entropy measures, namely sample entropy and approximate entropy were applied on the wavelet packet coefficients as nonlinear features, and significant features were selected via genetic algorithm (GA). Three machine-learning algorithms were used for classification; including support vector machine (SVM), multilayer perceptron (MLP) a novel enhanced K-nearest neighbors (E-KNN), which uses GA to optimize the feature-space distances and provides a feature importance index. The highest accuracy obtained by using frequency-based features was from gamma oscillations which resulted in 91.38%. Performance of nonlinear features were better compared to the frequency-based features and the results showed 94.28% accuracy. The combination of the features showed 98.44% accuracy with the new proposed E-KNN classifier.