Federated Learning for Privacy-Preserving AI in Healthcare
Hemanth Dandu · 2025
Through this study we investigate how Federated Learning (FL) maintains privacy for healthcare AI by utilizing actual hospital data which includes patient demographics and medical records. Research employed implementation and comparison of different machine learning algorithms including Logistic Regression, Random Forest, Naive Bayes, and Multi-Layer Perceptron (MLP) in the centralized and FL environments. The performance of the model was evaluated in terms of Accuracy, Precision, Recall, and F1 Score metrics and qualitative inspections including Confusion Matrices, ROC Curves, and Feature Importance plots. The research demonstrated that federated models had the same performance as centralized models while the Federated MLP model depicted better outcomes by higher AUC and balanced accuracy on all simulated clients. The findings demonstrate that FL offers a robust alternative to traditional centralized machine learning training approaches. FL enables secure model development along with data protection of the patient, which is very important for medical domains suffering from data sharing limitations, which in turn impedes AI deployment and interdisciplinary collaboration between greater than one organization.