Detection of Parkinson's disease Using Machine Learning Algorithms

Abstract

A Comparative Study of Support Vector Machine, Random Forest, and Gradient Boosting Classifier Parkinson disease (PD) is a neurodegenerative disorder that is disabling and non-motor, in which millions of people around the globe are affected. This paper constructs and analyzes the performance of the Machine Learning (ML) algorithms in PD detection using clinical data. On a dataset of 756 PD 564 patients and 192 healthy participants, we obtained a set of seven features on dataset, trained and tested Support Vector Machine (SVM) algorithm, Random Forest (RF) algorithm, and Gradient Boosting Classifier (GBC) algorithm. Findings indicate that GBC had maximum accuracy. The SVM also had a relatively poorer performance compared to GBC yet relatively good, Random Forest (RF) is also a good performer but scored a little low. Analysis of feature importance showed use of such as Pitch Period Entropy (PPE), Detrended Fluctuation Analysis (DFA), Recurrence Period Density Entropy (RPDE), Number of Pulses (numPulses), Number of Periods of Pulses (numPeriodsPulses), Local Shimmer (locShimmer) and Amplitude Perturbation Quotient 3 (apq3Shimmer) which contribute to a better detection of the Parkinson disease. The paper shows that ML algorithms are effective to identify PD and help the clinician to diagnose and plan treatment earlier, which will lead to the improvement of patient outcomes.