Feature Selection for Activity Classification and Dyskinesia Detection in Parkinson's Disease Patients
Abstract
Recent advances in wearable sensing technologies have favored the search for reliable and objective methods of estimating motor symptoms and complications of Parkinson's disease (PD). In this paper, we present a complete system of motor assessment composed of Shimmer3 inertial measurement modules aimed to classify a series of daily life activities performed by PD patients and detect the occurrence of Levodopa Induced Dyskinesia (LID). Feature selection methods are implemented on datasets collected from nine healthy individuals and 2 PD patients in order to determine the most relevant module positions with respect to activity classification and detection of LID. Classifying activities resulted in an overall accuracy of 88.05% in healthy individuals and 85.87% in PD patients, while detection of dyskinesia yielded 83.89%. The lowered performance is likely to be caused by the difficulty of classifying PD patients' activities due to presence of motor dysfunction.