The aim of this paper is to build a predictive model that can accurately predict the total UPDRS scores with at least 0.9 R Squared based on the voice measures. We utilized a dataset that recorded voice measures from 42 subjects with Parkinson Diseases, which consist of 5874 observations and 22 attributes. After performing data exploration, we noticed voice measures have a nonlinear relationship with UPDRS scores. Therefore, 3 nonlinear models are deployed, which are Support Vector Machine, Decision Tree Regressor, and Random Forest Regressor. After running all the models, the random forest regression model can best predict UPDRS based on voice measures (RQ2). This is because it has the highest R Squared (0.912), and has the lowest RMSE (3.39) and MAE (2.38) on the testing set. Besides, age is the highest feature importance in the prediction of UPDRS, which means that it is the most important predictor in the UPDRS score.
You can view the presentation via this link
You can view the competition via this link