Hybrid Deep Learning Framework for Adversarial Attack Prediction in Wireless Body Area Networks Using Spatiotemporal Data

In this work, we introduce a deep learning algorithm for WBAN adversarial attack prediction using spatiotemporal medical data of the MIMIC dataset that includes heart rate, blood pressure, body temperature, and oxygen consumption. We pair a 3D CNN innovation with BiLSTM because a 3D CNN can record patterns in space, and BiLSTMs do just that for time. After that, the preprocessing of the data with signal improvement and normalization of the data, the architecture consists of a convolutional and max pooling layer, and a BiLSTM layer to classify the data as normal or adversarial. Lastly, we could generate a matrix of experimental outcomes with high efficacy: accuracy (96.96%), precision (95.21%), recall (97.22%), F1 score (96.32%), detection rate (97.09%), and adversarial robustness (95.45%), which was higher than in conventional approaches. This study is in favor of WBAN security and therefore supports more patient safety and reliability in wearables.