Rényi Differential Privacy with a Temporal Convolutional Network for Privacy Preserving Data in Healthcare IoT

In modern era, the significant growth in the deployment of Internet of Healthcare Things (IoHT) devices has led to rise in concern about data privacy and security, particularly in Intrusion Detection System (IDS). However, existing Federated Learning (FL) based approach integrated, E-Differential Privacy (DP) with Convolutional Neural Network (CNN) and Deep Neural Network (DNN) but struggled with tradeoff between accuracy and privacy, lack of ability for temporal pattern recognition and limited adaptability to sequential healthcare data. Therefore, a novel privacy preserving FL framework that integrates Rényi Differential Privacy with a Temporal Convolutional Network (RDP-TCN) is proposed. Initially, the input data is collected from Washington University in St. Louis - Enhanced Healthcare Monitoring System (WUSTL-EHMS-2020) and East Carolina University - Internet of Healthcare Things (ECU-loHT) dataset. Next, the collected data is preprocessed by performing min max normalization and Synthetic Minority Oversampling Technique (SMOTE). After preprocessing, the input data is passed to the edge devices in FL architecture, where a TCN is trained by each client to capture long term dependencies. Further, RDP is used to process the trained local model gradients to ensure privacy preservation. Then, Federated Averaging is used by central server to aggregate the differentially private updates, thereby to obtain the global TCN model. Finally, the global TCN model performs intrusion detection across healthcare IoT environments. Therefore, the proposed RDP-TCN obtained 97.21% accuracy and 0.29 of privacy budget when compared to existing models.