Machine Learning Techniques for Automatic Depression Assessment

Abstract

Depression is one of the most common mood disorder that is inherently related to emotions, involving bad mood, low self-esteem and loss of interest in normal pleasurable activities. The aim of this work is to develop a framework based on the dataset provided by AVEC'14 for depression assessment. The proposed work presents two different motion representation methods: (a) Gabor Motion History Image (GMHI), and (b) Motion History Image (MHI). Several combinations of appearance-based low level features are extracted from both motion representations. These features were further combined with statistically derived features, and used for training and testing with several machine learning techniques. The proposed approach reached an F1 score of 81.93%, both for MHI and GMHI, with SVM classifier. The achieved performance is comparable to state-of-the-art approaches, while manages to outperform several others. Apart from accomplishing a competitive performance, the proposed work provides an exhaustive exploration of different combinations of the investigated motion representations, descriptors, and classifiers.