Anomaly Detection in Wireless Body Area Networks using Generative Adversarial Networks

The Wireless Body Area Network (WBAN) seam-lessly interconnects wearable devices on the human body, facilitating the monitoring of various physiological parameters for better health outcomes. Ensuring the accuracy and security of data transmission is paramount to prevent false alarms and maintain the overall reliability of WBANs. In response to these challenges, we propose a cutting-edge solution, the WBAN-GAN model. This innovative approach integrates a Generative Adversarial Network (GAN) to assist in the detection of anomalies within WBANs. We have employed a convolutional neural network with a lookback in the data to design our generator and discriminator to explore the spatiotemporal correlations in detecting anomalies. In this work, we fully utilize the outputs of both the generator and discriminator in identifying anomalies by calculating the anomaly score using reconstruction error and discriminator output. To the best of our knowledge, very little research has been done using GANs for anomaly detection in WBANs. To validate the efficacy of our WBAN-GAN model, we conducted extensive testing using the MIMIC dataset [11]. Comparative analysis with a CNN-Autoencoder-based anomaly detection model demonstrates the superior performance of our proposed approach. Our results show an accuracy of 97% in detecting the anomalies.