Cybersecurity Mechanism for Automatic Detection of IoT Intrusions Using Machine Learning

Abstract

This article proposes an ML-based cyber security mechanism to optimize intrusion detection that attacks internet objects (IoT). Our approach consists of bringing together several learning methods namely supervised learning, unsupervised learning and reinforcement learning within the same Canvas. The objective is to choose among them the most optimal for classifying and predicting attacks while minimizing the impact linked to the learning costs of these attacks. In our proposed model, we have used a modular design to facilitate the implementation of the intrusion detection engine. The first Meta-learning module is used to collect metadata related to existing algorithmic parameters and learning methods in ML. As for the second module, it allows the use of a cost-sensitive learning technique so that the model is informed of the cost of intrusion detection scenarios. Therefore, among the ML classification algorithms, we choose the one whose automatic learning of intrusions is the least expensive in terms of its speed and its quality in predicting reality. This will make it possible to control the level of acceptable risk in relation to the typology of cyber-attacks. We then simulated our solution using the Weka tool. This led to questionable results, which can be subject to the evaluation of model performance. These results show that the classification quality rate is 93.66% and the classification consistency rate is 0.882 (close to unit 1). This proves the accuracy and performance of the model.