Distributed Learning for Heart Disease Risk Prediction Based on Key Clinical Parameters with Evaluation Metrics Analysis

Abstract The purpose of this study design and test a Decentralized Federated learning framework that integrates a Mutual Learning approach with a Hierarchical Dirichlet Process-based Federated Learning Decentralized (HDP-FLDec) algorithm to predict heart disease. Conventional machine learning models in health care are data-centralized, which poses major risks to patient privacy and lacks support for handling data heterogeneity between institutions. Current federated models are also hard-pressed by non-IID data and lack scalability. Privacy protecting prediction of heart disease is for early treatment and intervention. A decentralized and intelligent federated method can improve diagnostic accuracy without compromising patient data, which is kept local and secure.A Decentralized Federated Learning framework, combining mutual learning between clients and employing HDP as a model for the latent structure of diverse health data.The endpoints were prediction correctness and AUC-ROC. Secondary endpoints were precision, recall, F1-score, and communication efficiency of the network.Heart disease data sets from public benchmark sources, namely UCI Heart Disease and Cleveland datasets of varying demographic and clinical records.Performance measures were calculated over several training in a decentralized peer-to-peer topology against centralized models and standard federated algorithms (FedAvg, FedProx).The HDP-FLDec model attained better prediction performance in all measures against baseline models.The model scored an accuracy of 94.2%, a precision of 92.8%, a recall of 93.6%, an F1-score of 93.2%, and an AUC-ROC of 0.97.HDP-FLDec beat baseline federated models by 5–7% accuracy and showed improved stability under non-IID data conditions.The mutual learning mechanism speed of convergence, while the HDP aspect delivered better personalization without compromising generalization. Communication was reduced by 18% compared to baseline federated architectures.The suggested HDP-FLDec algorithm provides privacy-preserving and accurate heart disease prediction in a decentralized environment, efficiently dealing with non-IID data and communication limitations.The model offers a practical solution for healthcare systems in real life to enhance diagnostic models simultaneously while preserving the confidentiality of patients. Integration with real-time monitoring systems and IoT-based medical devicesHDP-FLDec marries Bayesian nonparametrics' flexibility with the decentralization and privacy offered by federated learning while presenting greater precision, strength, and scalability of existing models.