Regularized CNN Based Model for Analyzing, Predicting Depression and Handling Overfitting

Abstract

Depression is serious and common disorder among human being that affects mental health. Large scale research is carried out to identify the risk of depression. Using technology to identify individuals with depression could connect patients with the help they need more quickly and easily while reducing healthcare costs and burden on physicians. In this paper Convolution Neural Network (CNN) is implemented using Patient Health Questionnaire (PHQ-9) screened NHANES (National Health and Nutrition Examination Survey) dataset from 1999 to 2014. The proposed framework for automatic depression prediction shows accuracy starting from 92.37 to 1.0. Continuing the research, dropout layer is used in visible and hidden layers to avoid overfitting. After the use of dropout layer the training and validation accuracy is synchronized and validation loss is less than training loss. Hence generalized model is obtained by using regularized CNN with 100% accuracy. When the implemented model is compared with existing work in the same area then it is observed that till date using CNN this is the first attempt to achieve 100% accuracy on NHANES dataset for identifying the risk of depression using PHQ-9 questionnaire.