Knowledge discovery, rehabilitation robotics, and serious games: Examining training data

Abstract

In this paper, we present an initial attempt to apply Knowledge Discovery techniques over real performance data from patients enrolled in robotic therapy in order to explore how to better optimize therapy. Performance data sets encompass measurements such as position, velocity and force, as well as final performance measures. We apply the Principal Component Analysis method in an attempt to reduce the dimensionality of the problem, molding subsets that were the input into a Multilayer Perceptron Artificial Neural Network which would carry out data mining with the purpose of discovering the relative significance of each field, in relation to a performance measure. It was possible to notice the impact caused by the lack of each field in terms of specific performance measures, indicating which data are more relevant to use in further experiments.