Journal of Internet of Things and Information Technology ISSN: 3107-4502 (Online)

Authors: M Naga Triveni, G Y Vybhavi, Eswar Poluri

Abstarct: It is possible that someday the Industrial Internet of Things will alter the planet. It is the amount of knowledge so far that lets the universe spin quicker. Detecting unexpected occurrences, adjustments or transitions in databases is one way to process data quickly and more effectively.  Anomaly detection, a technique that focuses on Artificial Intelligence to recognise irregular behaviour inside the data collection pool, has since become one of the Industrial IoT’s key goals. Anomaly detection relates to the discovery of objects or occurrences in a dataset that are normally undetectable by a human specialist that do not adhere to a predicted trend or to other items. Typically, those irregularities may be converted into concerns such as design flaws, mistakes, or theft. We recommend a two-phase model in this paper to identify and categorise irregularities. First, out of eleven widely used algorithms tested for the same data collection, we chose Random Forest based on the maximum accuracy-score. The RF is used to identify irregularities and create a “attack-or-not†additional function. We supplied the Neural Network with the “attack-or- not†data function to distinguish attack types, which will help to handle each form accordingly.

Keywords: Machine Learning, Neural Network, Cyber Security, Network Anomaly Detection and UNSW-NB15