A Chronic Psychiatric Disorder Detection Using Ensemble Classification

Abstract

The objective of this work is to detect depression that is a more chronic psychiatric disorder found in humans using speech samples. This work is first of its kind in depression detection using Audio Visual and Emotional Challenge 2011 (AVEC 2011) and Computational and Paralinguistics challenge 2016 (ComParE 2016) feature sets. A novel method of ensemble classification using simple machine learning algorithms of Instance-Based classifier with parameter K (IBK), Stochastic Gradient Descent (SGD) and Random Forest is proposed for the projected task with gender dependent and independent systems. Experimental results demonstrate the superiority of ComParE 2016 over AVEC 2011 in determining the psychological state of an individual. Feature selection method is applied to reduce feature vector size, maintaining the accuracy of depression detection with that obtained using large size feature sets. The ensemble-based classification provide better accuracy performance than the individual classifiers.