Blockchain-Governed Federated Learning with Sparse-Causal Bi- RNN Transformer Net for Secure and Intelligent Healthcare Analytics

Abstract To address the challenges of data privacy, temporal modeling inefficiencies, and lack of trust in centralized healthcare AI, this research proposes a novel Blockchain-Governed Federated Deep Learning Framework (B-FDLF). This framework integrates a hybrid architecture named Sparse-Causal Bi-RNN Transformer Net, designed for secure, scalable, and accurate prediction of neurodegenerative diseases like Parkinson’s. The model combines CNNs for local spatio-temporal feature extraction, Bidirectional RNNs for sequential learning, and a Sparse-Causal Transformer for long-range temporal dependency modeling with causal masking. To preserve patient privacy, federated learning enables local training at healthcare edge nodes, while blockchain integration ensures tamper-proof, auditable, and verifiable model updates via smart contracts. Experimental validation on the UCI Parkinson’s biomedical voice dataset demonstrates superior performance of the proposed method, achieving 94.8% accuracy, 93.9% precision, 94.3% recall, and 94.1% F1-score, while minimizing communication overhead (48.3 MB), privacy leakage (1.2%), and achieving high adversarial detection rate (96.4%). These results confirm that B-FDLF is a robust and privacy-preserving AI framework well-suited for real-world, decentralized healthcare applications.