Classifying Different Stages of Parkinson’s Disease Through Random Forests

Abstract

Parkinson’s disease (PD) is a progressive, neurodegenerative and age-related disease whose clinical characteristics include both motor and non-motor symptoms. Gait analysis, a three dimensional, non-invasive and computerized analysis of gait, can analyse walking features and carry out spatial and temporal parameters that can be included in machine learning algorithms. Knime analytics platform is employed to implement Random Forests. The aim of the present research is to distinguish De Novo PD patients (patients in early phase, without treatment) and Stable PD patients (patients in intermediate phase, in stable treatment) using spatial and temporal parameters of gait analysis. The dataset consists of 59 people, 32.2% De Novo and 67.8% Stable patients. Results show high accuracy (84.6%) and capacity to detect De Novo patients (94.9% of sensitivity). Recall and precision got high values, too. Despite needing further investigation, this pilot research should encourage health policy and facilities to introduce machine learning techniques and gait analysis in clinical practice. Moreover, results suggest the existence of gait patterns characterizing each phase of Parkinson’s disease.