Machine Learning for Clinical Score Prediction from Longitudinal Dataset: A Case Study on Parkinson’s Disease

Accurate prediction of Parkinson’s disease (PD) progression is vital for personalized treatment and effective clinical trials. This study presents a machine learning approach to predict the Movement Disorder Society-Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) Part III scores, quantifying motor symptom progression in PD patients. Using the longitudinal Parkinson’s Progression Markers Initiative (PPMI) dataset, we examined the impact of dataset format (wide vs. cross-sectional), dimensionality reduction techniques (PCA, NMF), and regression models (Linear Regression, Random Forest, XGBoost, SVR) on prediction performance. Our findings indicate that models trained on wide-format datasets consistently outperformed those on cross-sectional data. The combination of Nonnegative Matrix Factorization (NMF) and Support Vector Regression (SVR) achieved the best performance, with a mean absolute error (MAE) of 1.91 and R \(^{2}\) of 0.83. These results underscore the importance of data arrangement and highlight NMF’s effectiveness in feature extraction for longitudinal datasets.