CardioFL: A Federated Learning Prototype for Heart Disease Prediction Using Simulated Hospital Clients
Machine-learning models for heart-disease prediction are commonly trained on data collected in a single repository, which may raise data-governance and privacy concerns when multiple healthcare organizations are involved. This paper presents CardioFL, an educational federated-learning prototype that demonstrates collaborative Logistic Regression training across three simulated hospital clients using the Cleveland heart-disease dataset. The prototype implements local mini-batch training, weighted Federated Averaging (FedAvg), and a FedProx experiment for heterogeneous client data. In the centralized-versus-federated experiment (240 training and 57 test records), the centralized model achieved 85.96% accuracy (cross-entropy loss 0.3769) and the federated model 87.72% (loss 0.3884). In a separate experiment (238 training and 59 test records), IID + FedAvg, Non-IID + FedAvg, and Non-IID + FedProx achieved 84.75%, 81.36%, and 86.44% accuracy, respectively. A separate PyTorch reference experiment (237 training and 60 test records) reported 88.33% centralized and 85.00% federated accuracy. These experiments use different configurations and are not directly interchangeable. The three clients are simulated within one application backend; the prototype implements neither secure aggregation nor differential privacy and has not undergone clinical validation. The results demonstrate a federated-learning workflow on a small benchmark dataset, not clinical readiness or formal privacy protection.
Authors
- Rakesh Gowda D S
- Chaithanya M
- Guru Raghavendra D S
- Chandrashekarmurthy K M
- Manjunath N
Publication Details
- Journal
- International Journal of Science Strategic Management and Technology
- Published
- 2026-11-17
- DOI
- https://doi.org/10.55041/ijsmt.v2i11.008
- Primary Topic
- Privacy-Preserving Technologies in Data
- Type
- article
- Field-Weighted Citation Impact
- 0.00