This is an implementation of a privacy-preserving federated learning pipeline utilizing Dockerized nodes to simulate Fitbit users and the FedAvg algorithm. This involved training local models on patient data, including smartwatch readings such as BP, Oxygen Level, Stress Level, Activity Level, HR, and Sleep Duration, and then aggregating these updates into a global model using TensorFlow. Through iterative refinement, I successfully improved the overall model accuracy. The entire system was containerized using VS Code and Docker for efficient deployment and management. Dataset: Fitbit data from multiple users as follows
Link to project Expanation, Code and Implementation Steps https://docs.google.com/document/d/1IPKtMc8mEDwG_75HywXihW-k_Uhfj4AAVl_HIkoG_pY/edit?usp=sharing