Towards securing Wireless insulin pump system using unsupervised deep learning technique
M Shobana, S. Poonkuzhali · Research Square · 2022
Abstract With the advent of Internet of things (IoT) technology across various fields give arise to occurrence of many smart objects/things. These kind of smart objects involves even in medical area to achieve smart health care monitoring system, wearable devices, and medical implanted devices, in general those kind of systems were called as internet of medical things. Hence the tremendous increase in smart devices among medical domain has both pros and cons. Among many existing problems, security issues sounds to be most addressable problem in the IoT based medical application. These IoMT devices were considered to be resource constrained, so it does not possess enough security framework to fight against all sorts of malicious attack as well as data privacy of patients. The malicious attack in IoMT system can bring huge data loss and life threat to patients. The existing solutions suggested for IoMT security issues are relies on supervised learning, so this work is heavily based on unsupervised learning to improve the efficiency of the designed security model. In this paper, an intrusion detection system has been designed for most significant IoMT device namely Insulin pump system for diabetes treatment using deep learning technique in an unsupervised manner. In this model deep autoencoder has been utilized to classify the unauthorized insulin value from the legitimate insulin value and this model used insulin logs of several patients as its dataset. The performance of the designed model has been evaluated using the quality metrics like accuracy, precision, F1-measure, and recall. Furthermore the resultant model is compared and analyzed against existing methodology as well as traditional machine learning classifiers.