A Learning based Secure Anomaly Detection for Healthcare Applications

The wireless body sensor network (WBSN) is an emerging technology in the healthcare domain, which collects data from vital parameters of the patient's body, using small portable components such as sensors. It allows to continuously monitor patients without restricting their movements and to inform the health specialist of the evolution of their states. However, the deployment of new technologies in health applications without taking into account data security makes the privacy of patients vulnerable. Thus, these systems have insufficient performance for large and varied data sets, because they do not process information sufficiently diverse to cover all scenarios. Moreover, the interpretation of physiological signals is a tedious process in which human errors can be caused.To address these issues, we propose, in this paper, an effective health system that ensures the secure collection and transfer of patient data to the physician, to facilitate the diagnosis and the detection of life threatening diseases. Our main contribution is to offer real-time, high quality processing, learning and analysis capabilities, in a smart and secure system. For that, we have collected real patient datasets from Tunisian doctors.To achieve this goal, we used ANN and XgBoost as learning algorithms and AES as an encryption protocol for sending secure data to medical personnel for diagnostic purposes. The results show an accuracy of 80% with an execution time of 1.54 s.