Enhancing security for smart healthcare in wireless body area networks using a novel adversarial detection using ACR BiLSTM with multi-batch stochastic gradient descent

Wireless body area networks (WBANs) are essential for intelligent healthcare services as it monitors the patient data in real time. The confidentiality of sensitive medical records is at risk since WBANs are vulnerable to potential attacks from malicious actors and hence the protection of sensitive healthcare data is of extreme importance. The vulnerability in WBANs to malicious operations poses a substantial threat to the dependability and security of intelligent healthcare services. Such undisputable need exists for robust algorithms that are capable of identifying and effectively defend hostile attacks. Thus, in real-time WBAN data, this study develops a novel method for identifying potential threats by developing an Attention Convolutional Residual Bi-LSTM (ACR BiLSTM) approach that classifies the adversarial from WBAN real-time data. This research evaluates the robustness of ACR BiLSTM learning model against the adversarial inputs by comparing its error margin with the decision boundary. The utilization of Mini-Batch Stochastic Gradient Descent (MBSGD) generates adversarial samples while maximizing the loss function by considering a noise upper bound. Further, a Soft Margin Adversarial Training (SMAT) algorithm optimizes the adversarial noise upper bounds using the shortest possible distance between decision boundaries and clean samples. To achieve the highest possible accuracy, the proposed SMAT modifies the sample margins dynamically during training. By positioning adversarial training examples in closer proximity to decision boundaries, this proposed method enhances the robustness without compromising the accuracy on clean data. For training and validation, the research uses MIMIC II databases for classification of signals and the feature extraction and adversarial classification. The accuracy, precision, recall, F-measure, and Area Under the Curve shows that the ACR Bi-LSTM algorithm is identifies the adversarial attacks in WBANs than the state-of-art techniques.