Privacy-Preserving Artificial Intelligence for Diabetes Prediction: A Comparison of Centralised and Federated Learning

Abstract

Artificial intelligence (AI) is increasingly used in healthcare to support disease prediction and clinical decision-making. Traditional centralised machine learning approaches often require the aggregation of sensitive patient data into a single repository, which raises substantial privacy, ethical, and regulatory concerns. Federated learning has emerged as a privacy-preserving alternative that enables collaborative model training across distributed data sources without sharing raw patient data. In this study, we investigate whether federated learning can achieve predictive performance comparable to that of centralised machine learning when applied to structured healthcare data. Using the PIMA Indians Diabetes dataset, several centralised machine learning models are evaluated and compared with a federated logistic regression model implemented using the Flower framework. Model performance is assessed using standard classification evaluation metrics. The results show that the federated approach achieves performance comparable to the centralised baselines, with a small reduction in predictive performance. These findings indicate that federated learning is a practical and effective solution for privacy-preserving predictive modelling in healthcare.

Ähnliche Arbeiten