This project investigates a feasibility of diagnosing patients with Parkinson’s disease from the motion data. A goal is to create a video-based AI model that would facilitate early detection of Parkinson’s disease with objective diagnosis data. It explores various machine-learning methods, including Gaussian model, FDA, SVM, Gaussian process, MLP and HMM (Fig. 5). While most methods could discriminate patients from healthy control using motion video data, a relatively small number of patients’ data poses necessity to collect more patients’ data as well as a means to deal with data imbalance issues.