A machine learning-based efficient anomaly detection system for enhanced security in compromised and maligned IoT Networks

Abstract

The proliferation of the Internet of Things (IoT) has brought about unprecedented connectivity and automation, yet it has also introduced significant security vulnerabilities. Malicious actors exploit these vulnerabilities, posing a grave threat to IoT networks and their users. Traditional machine learning approaches cannot detect these threats because IoT data is complex. In this paper, we introduce a novel anomaly detection framework specifically designed for compromised and maligned IoT environments. The proposed approach combines Modified Whale Transfer and Sine-Cosine algorithms along with feature selection techniques such as ANOVA, RFE, and RFA to detect malicious communications accurately. We use HP Tuned Machine Learning Algorithms further for developing an anomaly detection model and optimization of them by implementing machine learning algorithms. Our proposed approach is evaluated on UNSW NB-15 dataset, which enables characterization of anomaly detection very precisely. In our framework, diversified performance metrics and extensive experimentation show the promise of results in effectively detecting malicious activities in IoT networks. The model exhibits high performance in the case of binary classification with 99.9 % accuracy. For multiclass metrics like F1-Score, precision and recall outperform with the 94.36%, 94.81% and 93.92% respectively using UNSW NB-15 database.

Topics & Concepts