Exploring Machine Learning to Analyze Parkinson's Disease Patients

Abstract

Parkinson's disease (PD) is the second most common neurodegenerative disorder. Changes in gait kinematics and its spatiotemporal features are hallmarks for the diagnosis of PD. Lower limbs movement analysis is intricate and usually requires a gait and biomechanics laboratory; these complex systems are not always available in the medical consultation. This paper evaluates and proposes a machine learning classifier for the analysis of people diagnosed with PD through their gait information. This model has an accuracy of 82%, a false negative rate of 23% and a false positive rate of 12%, results were obtained from a training process that incorporates a low cost system that uses RGBD cameras (MS Kinect) as the main motion capture and the best features detected during an exploratory data analysis. Our study was evaluated using data harvested through the system mentioned and measurements from 60 volunteers; there were 30 subjects with PD and 30 healthy subjects.