Machine Learning for Affective Computing: Challenges and Opportunities

This chapter is from the forthcoming The Oxford Handbook of Affective Computing edited by Rafael Calvo, Sidney K. D’Mello, Jonathan Gratch, and Arvid Kappas. The ability to recognize affect is one of the fundamental requirements in building a computerized affectively intelligent system. Although the ideas from traditional machine learning constitute the core of an affect recognition methodology, there are significant additional considerations that must be observed. These considerations include aspects of data collection and annotation, feature selection, algorithm design, and system evaluation. This chapter aims to highlight such deviations and provide an overview of how some of the current research has attempted to solve these problems.