Classifying Elderly Emotion

Abstract

Classifying human emotion has been a research field of interest for quite some time now. This research will add to existing work on the subject, and look into the classification of emotion of elderly people. As a part of the 2020 ComParE paralinguistics sub-challenge on elderly emotion, I will conduct experiments with different machine learning models, feature representations, principle component analysis and feature level fusion, to find which combination leads to the best results on this classification task. The support vector machine classifier shows promising results on the arousal classification task, and feature level fusion also shows promising results on the valence classification task. The application of principle component analysis shows mixed results, some positive and some negative. Other researchers, as part of the team working on the sub-challenge, might benefit from these results, but should be cautious to apply them in all cases.