Anomaly Detection in WBANs using GANs and ConvLSTM

Since abnormalities in physiological signals might be a symptom of major medical disorders or device failures, anomaly detection in Wireless Body Area Networks (WBANs) is crucial for accurate and trustworthy health monitoring. To address the distinct spatiotemporal features of WBAN data, this research proposes a novel method called the WGAN-ConvLSTM model. Our model effectively captures spatial and temporal characteristics in physiological parameters collected from human body by combining Long Short-Term Memory (LSTM) layers with Convolutional Neural Networks (CNNs) in the encoder-decoder structure. Furthermore, we use Wasserstein loss to improve model robustness and training stability, resolving conventional GAN-based anomaly detection issues. The WGAN-ConvLSTM model was tested against several deep learning architectures frequently employed in anomaly detection, such as regular GANs, CNNs, and LSTM networks. Our model showed exceptional accuracy and robustness across several key performance parameters, with an Accuracy of 99.17%, Recall of 98%, ROC-AUC Score of 99%. A crucial component for real-time health monitoring applications, the model’s excellent recall, and ROC highlight its ability to identify anomalous trends while reducing false negatives. The WGAN-ConvLSTM model performs noticeably better than current models in identifying intricate, context-dependent abnormalities in WBAN data, which makes it a viable option for medical applications that need precise, real-time anomaly detection models.