Exploratory Boosted Feature Selection and Neural Network Framework for Depression Classification

Abstract

Depression is a burdensome psychiatric disease common in low and middle income countries causing disability, morbidity and mortality in late life. In this study, we demonstrate a novel approach for detection of depression using clinical data obtained from the on-going Mysore Studies of Natal effects on Ageing and Health (MYNAH), in South India where the members have undergone a comprehensive assessment for cognitive function, mental health and cardiometabolic disorders. The proposed model is developed using machine learning approach for classification of depression using Meta-Cognitive Neural Network (McNN) classifier with Projection-based learning (PBL) to address the self-regulating principles like how, what and when to learn. XGBoost is used for feature selection on the available data of assessments with improved confidence. To improve the efficiency of McNN-PBL classifier the best parameters are found using Particle Swarm Optimization (PSO) algorithm. The results indicate that the McNNPBL classifier selects appropriate records to learn and remove repetitive records which improve the generalization performance. The study helps the clinician to identify the best parameters to analyze the patient.