Deep Learning Approach for Incongruous Data: Dataset Matters

Today, humans seek opportunities to grow professionally and improve their mental health. As a result, anxiety and depression are now prominent to the point that people often accept them as part of their usual regular life routine. According to the World Health Organization (WHO), more than 300 million people around the world are suffering from depression and anxiety disorder. Therefore, being able to detect risk factors that lead to depression can be beneficial to the healthcare system. We use machine learning techniques to detect depression risk factors using a medical survey dataset, the Behavioral Risk Factor Surveillance System (BRFSS), available from the Centers for Disease Control and Prevention (CDC). This survey is obtained explicitly through telephone calls from adults 18 years and older who live in the United States of America. Oversampling and undersampling techniques and eight machine learning techniques are used to evaluate further and examine the data set. Among all, the SMOTE oversampling techniques show enhanced performance over the undersampling and no sampling techniques, and the deep neural network generates the best outcome compared to other techniques.