A Rule-Based Synthetic Multi-Domain Healthcare Dataset

Abstract

CardioHealth-FL: Synthetic Cardiovascular Healthcare Dataset CardioHealth-FL is a rule-based synthetic healthcare dataset developed for cardiovascular risk and federated learning research. It contains 40,000 records, 16 features, and a target variable, Cardiovascular_Risk. The records were generated using predefined probabilistic rules to represent different cardiovascular risk conditions. The target classes are Low (38.0%), Moderate (34.0%), and High (28.0%). The dataset can be used for healthcare analytics, federated learning, client heterogeneity, and robust learning experiments. DiaMet-FL: Synthetic Metabolic Health Dataset DiaMet-FL is a rule-based synthetic dataset designed to represent different metabolic health conditions for healthcare and federated learning research. It contains 40,000 records, 14 features, and the target variable Metabolic_Risk. The data were generated using predefined probabilistic rules, producing three classes: Normal (51.06%), Pre-Diabetic (33.79%), and Diabetic (15.14%). The dataset supports research on healthcare analytics, non-IID data, federated learning, and robust model development. MaternalHealth-FL: Synthetic Maternal Healthcare Dataset MaternalHealth-FL is a rule-based synthetic dataset developed to represent maternal health conditions in federated healthcare research. It contains 40,000 records and 22 features, with Maternal_Risk as the target variable. The records were generated using predefined probabilistic rules covering four risk levels: Low (54.21%), Moderate (31.05%), High (12.47%), and Critical (2.28%). The dataset can be used for maternal health analytics, federated learning, client heterogeneity, and trustworthy healthcare AI research. WearHealth-FL: Synthetic Wearable Healthcare Dataset WearHealth-FL is a rule-based synthetic wearable healthcare dataset created for research in healthcare analytics and federated learning. It contains 40,000 records, 20 health-related features, and the target variable Health_Status. The data were generated using predefined probabilistic rules to represent three health conditions: At_Risk (43.43%), Stable (28.52%), and Critical (28.05%). The dataset can support research involving wearable health data, heterogeneous federated clients, and robust learning under dynamic conditions. All four datasets are synthetic and rule-based. They were generated using predefined probabilistic rules and controlled distributions to create realistic variations in healthcare attributes and target classes. No real patient records were used, and the datasets were not generated using artificial intelligence or machine learning models. They are intended for research, experimentation, benchmarking, and educational purposes.