Enjoyment recognition from physiological data in a car racing game

Abstract

In this paper we present a case study on The Open Racing Car Simulator (TORCS) video game with the aim of developing a classifier to recognize user enjoyment from physiological signals. Three classes of enjoyment, derived from pairwise comparison of different races, are considered for classification; impact of artifact reduction, normalization and feature selection is studied; results from a protocol involving 75 gamers are discussed. The best model, obtained by taking into account a subset of features derived from physiological signals (selected by a genetic algorithm), is able to correctly classify 3 levels of enjoyment with a correct classification rate of 57%.