Decentralized Privacy-Preserving Framework for Sleep Anomaly Detection and Behavioral Pattern Recognition in Healthcare

Sleep is a fundamental biological need that is crucial for physical, cognitive and mental health. Lack of sleep affects daily activities, reduces productivity and increases the risk of stress, anxiety, depression and burnout. In healthcare, sleep-related data is analyzed using machine learning to understand behavioral patterns and predict health status. However, data privacy and security is a major challenge due to the use of sensitive personal data. To address this problem, Federated Learning (FL) ensures privacy by training models in a distributed environment, instead of storing data centrally. Explainable AI (XAI) enhances transparency and trustworthiness by explaining model decisions. Blockchain provides secure, immutable and tamper-proof storage of data. In combination, FL, XAI and Blockchain create a secure and privacy-preserving healthcare infrastructure. In this study, a decentralized framework is proposed by integrating Federated Learning (FL), Explainable AI (XAI), Blockchain Zero-Knowledge Proofs (ZKPs) and IPFS. The primary goal of this framework is to ensure safe and privacy-preserving sleep related data analysis in a distributed environment. For validating the proposed framework, an Autoencoder is used for sleep anomaly detection and four deep learning models are evaluated with eight FL algorithms along with XAI. In the performance results, the on-chain process (Blockchain, ZKPs, IPFS) showed high efficiency (22 TPS, 73 ms latency, 100% success rate, block confirmation time 0.2ms and average transaction time 0.07 sec), and in the off-chain process achieved 0.9776 accuracy and 0.9831 F1-score in the best FL-XAI setup (LSTM-Attention with pFedMe) and 0.9223 accuracy and 0.9337 F1-score in validation.