MACHINE LEARNING-DRIVEN INTRUSION DETECTION IN IOT VULNERABLE ENVIRONMENTS

Abstract : The method for the development of an advanced intrusion detection system specifically designed for the Internet of Things. In particular, we employ a deep-learning algorithm to identify malicious activity in Internet of Things networks. The detection solution makes a variety of network communication protocols used in the Internet of Things interoperable and offers security as a service. We assess the scalability of our suggested detection framework through simulation as well as real-network traces, which serve as a proof of concept. With 97% accuracy, 90.5% Matthews correlation coefficient (MCC), and 99.6% Area under the Curve (AUC) performance, XGBoost is the best supervised algorithm. Furthermore, a noteworthy discovery of this work is that the Expectation-Maximization (EM) algorithm, an unsupervised technique, beats other methods in the NSL-KDD dataset and likewise does rather well in the detection of attacks. This dataset, which is divided into nine distinct assault types, combines modern attacks with typical network traffic behaviors. Principal Component Analysis was used in the following step to reduce dimensionality (PCA). The validation dataset, accuracy, area under the curve, recall, F1, precision, kappa, and Mathew correlation coefficient (MCC) were used to assess the experimental results of our findings.