A Convolutional Neural Network Study on Depression and Eye Blink Analysis

Abstract

This study is determining the correlation of human blinks relating to depression. The study uses convolutional neural network for detecting blinks in a video. Using Closed Eyes in the Wild dataset the Convolution Neural Network model was trained having 99.24% in training accuracy and 0.0275 loss from epoch of 50. However, the results from validation of the model resulted 61.09% tested from two datasets that where labelled with BDI-II depression scale. The study collated the results of recorded blinks from the video datasets and it showed that there is a weak positive correlation of the recorded blinks computed as blinking rates to depression. The result showed that the r2 score was 0>3.4 thus, there is a possibility but not the highly indicator of depression.