Through the Internet of Things (IoT) healthcare professionals get access to fundamental clinical data which aids in early disease diagnosis and medical research development. Studied research indicates that Blockchain with Federated Learning establishes decentralized solutions for protected data storage and privacy-protecting machine learning processes. Current healthcare institutions avoid adopting IoT technology because of security threats alongside privacy concerns. The data security of centralized machine learning systems requires strong defensive measures because cyber attacks continue to threaten patient information safety. The research creates a diabetes prediction model through the combination of privacy-optimized systems which blend Federated Learning techniques with Blockchain structure. The XGBoost gradient-boosting algorithm used in the system delivers superior performance than standard Multilayer Perceptron (MLP) models by reaching 93% maximum accuracy level. Using the proposed Blockchain-FL-XGBoost framework delivers decentralized privacy protection with 97.3% accuracy alongside centralized 99.5% accuracy while maintaining minimal operational trade-offs. XGBoost brings stronger advantages to medical diagnostic applications running with Federated Learning compared to using traditional MLP-based approaches. The application of Blockchain technology alongside this secure AI model improves predictive accuracy and sets advanced secure standards that will drive healthcare AI systems based on privacy protection.