Predicting Depression Levels using Back Propagation Neural Network

Abstract

Depression is a mood disorder characterized by feelings of deep sadness and a sense of indifference. When depression recurs in moderate or severe intensity, it can be a serious health condition. The most effective way to deal with this problem is to predict the symptoms of depression at an early stage. In this study, a Back Propagation Neural Network (BPNN) model is proposed to predict whether a person is categorized as mild, moderate, or severe depression based on Beck's Depression Inventory (BDI) data. There are 21 BDI data items that are used as predictors in the built BPNN model. The dataset used is 227 patients and divided into 2 categories, namely 181 observations (80%) as training data and 46 observations (20%) as test data. While the BPNN model that is built has 21 neurons in the input layer, one hidden layer with 21 hidden neurons and 4 neurons in the output layer. After testing, it was found that the BPNN model is able to predict the level of depression with F1-Score of 100, 95.65, 90.91 and 95.24% for the classification of normal, mild depression, moderate depression and severe depression, respectively. Overall, the accuracy level reached 95.65%. This study concluded that the proposed model can help doctors or psychiatrists to predict depression at an early stage, whether it is classified as mild, moderate, or severe depression, so that the patient can receive appropriate treatment.