A Home-Based Diabetes Prediction System on Internet of Things, Federated Learning and Edge Computing

Detecting the disease early is an important step in reducing its impact. In recent years, applications that monitor and predict health metrics using machine learning have attracted public attention. Our research built a diabetes health monitoring and prediction system based on the Edge Computing model. For hospitals, users and patients are represented by K3 clusters. The K-Nearest Neighbor (KNN) algorithm is run in a distributed fashion using Federated Learning with the proposed system. It could allow people to track their vital health indicators without having to go to the hospital. In our proposed system, the diabetes risk level can be predicted in advance so that users can take preventive steps. Through Federated Learning used, the model is being trained on distributed data sources guaranteed to preserve privacy and improve accuracy. K-Nearest Neighbor in the federated learning cluster, we can improve the prediction accuracy by up to 10% compared to the standalone version of K-Nearest Neighbor.