An automated mechanism to characterize wheelchair user performance

This paper proposes a mechanism to derive quantitative descriptions of wheelchair usage as a tool to aid Occupational Therapist with their performance assesment of mobility platform users. This is accomplished by analysing data computed from a standalone sensor package fitted on an wheelchair platform. This work builds upon previous propositions where parameters that could assist in the assessment were recommended to the authors by a qualified occupational therapist (OT). In the current scheme however the task-specific parameters that may provide the most relevant user information for the assessment are automatically revealed through a machine learning approach. Data mining techniques are used to reveal the most informative parameters, and results from three typical classifiers are presented based on learnings from manual labelling of the training data. Trials conducted by healthy volunteers gave classifications with an 81% success rate using a Random Forest classifier, a promising outcome that sets the scene for a potential clinical trial with a larger user pool.