Ai-Driven Multi-Disease Prediction System Using Federated Learning For Privacy-Preserving Healthcare Analytics

Abstract

Medical institutions accumulate large volumes of patient data, yet strict privacy legislation and organizational constraints limit the exchange of such data between healthcare providers. These restrictions often prevent the development of robust machine learning models for clinical decision support. To address this challenge, this study introduces a Federated Learning (FL) based multi-disease prediction model that allows multiple hospitals to collaboratively train a global classifier without exposing sensitive patient information. Three major chronic conditions—diabetes, heart disease, and chronic kidney disease (CKD)—were selected to build a unified multi-class prediction system. Each participating institution trains the model locally, and the global model is updated using the Federated Averaging (FedAvg) approach. Classical machine learning algorithms and a deep neural network (DNN) were evaluated under both centralized and federated setups. The federated DNN achieved an accuracy of 92.4%, which is comparable to the centralized accuracy of 93.1%, demonstrating that high performance can be achieved without data centralization. The findings confirm that FL is a viable solution for privacy-aware multi-disease diagnosis and can be deployed in real-time healthcare analytics