Dytokinesis: A Cytokinesis-Inspired Anomaly Detection Technique for IoT Devices

IoT devices are becoming ubiquitous and the availability of open-source botnets has made it very easy for anyone to attack and manipulate such connected devices and even infect them. These anomalies are getting sophisticated and powerful enough to generate network traffic at terabits per second (Tbps) and cost companies over a billion dollars a year. We present a novel technique, named Dytokinesis, to separate such anomalous entities. Dytokinesis is inspired by the biological Cytokinesis process in which a cell is divided into two. Dytokinesis, on a similar pattern, performs such a division on a dataset with high accuracy and low latency. Dytokinesis works in different phases and makes use of Empirical Data Analysis (EDA) and Gaussian kernel to bisect the dataset into normal and anomalous classes. Experimental results demonstrate that Dytokinesis obtains significantly higher accuracy compared to other state-of-the-art techniques while achieving the best run-time performance.