Classification of Physiological Data in Affective Exergames

In this work, we present our approach to analyze physiological data in affective exergames by using deep learning algorithms. In previous works, we enhanced a cycling exercise machine to act as a game controller. During a case study, we then collected vision-based and physiological data of 25 participants who rode through a game environment that was designed to provoke emotions. In order to analyze the collected physiological data, we now propose an ensemble learning approach based on three distinct deep learning models: Multilayer Perceptron, Fully Convolutional Networks and Residual Networks. As a result, the proposed algorithms were able to enhance the quality of our event-based emotion analysis method introduced previously.