Statistical Analysis of Joystick Trajectories

Abstract

The research field of affective computing aims to improve human-machine interaction. One of the main goals is to enable autonomous systems to recognize and adapt to human emotions. Machine learning is able to find an attribution between physiological reactions and underlying emotions. In order to provide labelled training data, human subjects annotate emotional stimuli in experimental studies. Three major challenges of the resulting continuous annotation data are: 1. Finding a suitable representation of this complex data, 2. Comparing the annotations of different subjects, 3. Combining the annotations to provide reliable ground truth for machine learning. Since previous research did not take into account the continuous nature of the annotation data, a functional data approach is introduced: Annotations are represented as smooth functions in a low-dimensional functional eigenspace. Comparison and ground truth estimation is then performed using simple statistical methods.