Applying Federated Learning in the detection of Freezing of Gait in Parkinson’s disease

Freezing of Gait (FoG) is a motor symptom of Parkinson's disease, which causes an episodic inability to move in patients, negatively affecting their daily activities. So, it is vital to monitor and alert the FoG manifestation to help these patients. This study considers two major constraints for developing a healthcare application for FoG: the difficulty of collecting enough representative data and the privacy of the data collected from these participants. Therefore, we propose a Federated Learning (FL) healthcare application for wearable devices to detect FoG symptoms. We evaluate and compare the proposed model to a centralized machine learning approach. We employed a dataset with imbalanced classes of 10 patients with PD to train and test both models. The results show that the accuracy differs by just 1% from that of the centralized model and by 5% from when using the imbalanced training subsets after applying the SMOTETomek's balanced technique.