Differential Privacy Enabled Healthcare AI for Thyroid Disease Prediction

The increasing reliance on machine learning in healthcare, finance, and social media raises serious concerns about the protection of sensitive user data. Protecting patient data privacy while harnessing the power of artificial intelligence for disease prediction is a critical challenge in modern healthcare. Traditional machine learning models are vulnerable to privacy attacks such as membership inference and model inversion, which can expose individual records. To address this challenge, we propose a differential privacy enabled healthcare AI system for thyroid disease prediction, ensuring strong privacy guarantees without compromising diagnostic accuracy. The system employs Laplace noise for sensitivity-based perturbations, Gaussian noise for advanced composition scenarios, and exponential mechanisms for discrete optimization tasks. Experiments were conducted on thyroid datasets with privacy budgets (ε) ranging from 0.1 to 1.0 to systematically evaluate privacy-utility tradeoffs and analyze their effects on performance metrics such as accuracy, precision, recall, and F1-score. The results demonstrate that differential privacy can be effectively integrated into healthcare AI pipelines with controlled accuracy degradation, thereby fostering the development of secure, trustworthy, and privacy-preserving predictive systems for sensitive medical applications.