Parkinson's Disease Severity Assessment: A Comparative Study & Interpretability Analysis

Parkinson's disease (PD) is a progressive neurological disorder that mainly affects people over the age of sixty. Currently, there is no such cure for PD. However, Treatments are available to help relieve the symptoms related to PD. A rating tool called MDS-UPDRS III is used to gauge the severity and progression and stages of disease severity are measured by the Hoehn and Yahr (NHY) staging system. This study performs a comparative analysis of the effectiveness of various machine learning models such as AdaBoost, XGBoost, Gradient Boosting Classifier (GBC), Decision Trees (DT), K-Nearest Neighbor (KNN), Logistic regression (LR), and Gaussian Naïve Bayes (GNB) on a dataset containing a variety of PD-related features obtained from 2861 persons. Local Interpretable Model-agnostic Explanations (LIME) is employed to perform the analysis of the strength of each feature toward the outcome prediction for various classification tasks. It is observed that the AdaBoost classifier delivers superior performance of 93.2% accuracy and 84.2% F1-Score for the dataset.