Performance analysis of deep learning CNN in classification of depression EEG signals

With the advent of greater computing power each year, computer-based disease/condition diagnosis have been gaining significant importance recently. In this paper, an extensive analysis of the approach based on the classification of depression using electroencephalogram (EEG) signals is carried out. A computer-aided machine learning approach: Convolutional Neural Network (CNN), a deep learning method is used in this work. The deep CNN was trained using EEG signals from 30 normal and 30 depressed persons. The network attained the highest accuracy of 99.31% in classifying depression from EEG signals of normal controls recorded from the right hemisphere of the brain and 96.3% from the left hemisphere of the brain after ten-fold cross-validation. The performance of the CNN network was evaluated by evaluating the classification accuracy, varying different parameters such as the number of strides, learning rate parameter, number of epochs, and sample size. An extensive data learning approach is proposed to classify depression EEG signals from that of healthy controls. The key advantage of using deep learning is that they return state-of-the-art accuracy and do not require manual pre-processing or feature extraction from the signal.