A Proposed Intrusion Detection Method Based on Machine Learning Used for Internet of Things Systems

Abstract

This paper presents an improved method using supervised machine-learning techniques of the Internet of things (IoT) systems to ensure security in deployments devices. The method increases accuracy and efficiency, identifies patterns, and makes decisions with significantly reduced error. In this work, we compare previous works by our improved ML method for both binary and multi-class classification on some IoT datasets. Based on metric parameters such as accuracy, precision, recall, F1 score, and ROC-AUC, the simulation results reveal that Classification and Regression Trees (CART) outperforms on all types of attacks in binary classification with an accuracy of 99% and with an accuracy between 21% and 37% higher than the original one. However, in multi-class classification, Naive Bayes (NB) outperforms other ML algorithms with an accuracy of 99% and an accuracy between 1% and 4% higher than the others works.