Utilizing Ensemble Learning in Detecting Parkinson's Disease with Reduced Facial Expressions and Hand-Written Drawings

Parkinson's Disease (PD) is a neurodegenerative disorder that affects individuals' movement, planning, initiation, and execution. Some of the most common symptoms of PD are tremors, rigidity, and initiation difficulties. Due to the rigidity in facial muscles, PD patients experience reduced facial expressions also known as Hypomimia. And tremor effects on the PD individual's handwriting abilities. Therefore, these two symptoms can be used in detecting PD in the early stages. The utilization of computer-aided approaches in PD diagnosis has improved worldwide over the last few years. Among these techniques, Machine Learning (ML) models have proven a promising success in PD diagnosis. And it resolves most of the issues such as patients' participation in multiple tests for the diagnosis. In this study, video based facial expression dataset and handwriting drawing dataset are collected from PD diagnosed individuals and non-PD individuals as the input. The accuracy is measured with different algorithms. By applying two-layer stacking ensemble learning framework the predictive accuracy is enhanced. The ensemble makes use of the base models, an attention layer, and a stacking model to combine their predictive strengths. The base models consist of the Logistic Regression method and SVM model.