A privacy-aware healthcare framework with model pattern-deviation detection for heart-disease prediction using L2-GNAE and PDDP

Abstract Sensitive patient data protection is essential to ensure medical reliability in healthcare. Yet, the traditional studies didn’t analyze the deviation in the shared model pattern, thus resulting in poor diagnosis. Therefore, this article proposes a privacy-aware healthcare framework with model pattern deviation detection for Heart Disease (HD) prediction using L2 Gini Norm Auto-Encoder (L2-GNAE) and Polynomial Differential Decay Privacy (PDDP). Firstly, the patients are registered into the healthcare applications, followed by data sensing, data encryption, and hash code generation. Meanwhile, to authenticate the data integrity, the data decryption and hash code verification are done. During testing, the verified data is subjected to the trained proposed local model for HD prediction. Next, to perform model privacy, PDDP is used. Afterward, to effectively classify the HD, the Triple Gated Lipschitz Recurrent Unit (TGLRU) is utilized. Also, the local model gradients are updated in the global model, where L2-GNAE is utilized to detect the deviations in the shared model pattern. If the deviation is detected, then the alert is sent to the hospital; otherwise, the model update is carried out. The experimental testing of the proposed framework is done by using the “Heart Disease Prediction Dataset”. From the validation, the proposed framework achieves 99.2145% accuracy, 99.0237% precision, and 99.1046% F-measure during HD prediction. Thus, the proposed work significantly outperforms the traditional works by obtaining a high security level (256 bits) with enhanced privacy-preserved HD prediction in healthcare maintenance.