Optimized Association Rule Mining for Predicting Heart Diseases from Distributed Medical Databases

This study introduces a privacy-preserving technique for heart disease forecasting within the federated learning environment over distributed medical datasets. Local SVM models are learnt independently at each institution to preserve data privacy, but the aggregated insights are used with a global XGBoost model designed to counter class imbalance and improve predictive accuracy. Normalization and up sampling were used as pre-processing steps on data collected to enhance smooth training of the models. The model achieved superior performance, which shows 98.5% accuracy, a precision of 95.8%, and a recall of 93.6% in comparison with local SVM models and traditional association rule mining techniques.