Efficient, Lightweight Cyber Intrusion Detection System for IoT Ecosystems Using MI2G Algorithm

Abstract

The increase in internet connectivity has led to an increased usage of the Internet of Things (IoT) and devices on the internet. These IoT devices are becoming the backbone of Industry 4.0. The dependence on IoT devices has made them vulnerable to cyber-attacks. IoT devices are often deployed in harsh conditions, challenged with less computational costs, and starved with energy. All these limitations make it tough to deploy accurate intrusion detection systems (IDSs) in IoT devices and make the critical IoT ecosystem more susceptible to cyber-attacks. A new lightweight IDS and a novel feature selection algorithm are introduced in this paper to overcome the challenges of computational cost and accuracy. The proposed algorithm is based on the Information Theory models to select the feature with high statistical dependence and entropy reduction in the dataset. This feature selection algorithm also showed an increase in performance parameters and a reduction in training time of 27–63% with different classifiers. The proposed IDS with the algorithm showed accuracy, Precision, Recall, and F1-Score of more than 99% when tested with the CICIDS2018 dataset. The proposed IDS is competitive in accuracy, Precision, Recall, and training time compared to the latest published research. The proposed IDS showed consistent performance on the UNSWNB15 dataset.

An Improved Mutual Information Feature Selection Technique for Intrusion Detection Systems in the Internet of Medical Things

“…Figure 3 , Figure 4 , Figure 5 , Figure 6 , Figure 7 and Figure 8 show comparisons between the accuracy achieved by the proposed LRGU-MIFS and the techniques proposed by [ 44 , 48 , 57 ]. Moreover, they show accuracy comparisons between the LRGU-MIFS and the related studies using feature sets with sizes 5, 10, 15, 20, 25, and 30, respectively.…”

“…This enhanced capability ultimately leads to more accurate feature selection and an improved overall performance. It is worth noting that the highest accuracy average was obtained by the DT classifier for the features selected by the proposed LRGU-MIFS and RCGU-MIFS [ 48 ], as opposed to the other techniques [ 44 , 57 ] that did not follow the redundancy gradual upweighting approach. This can be attributed to the low redundancy of the selected features, which makes the DT able to identify the attack patterns more clearly based on those features.…”

“… Comparisons of the average accuracy between the proposed LRGU-MIFS and related works (Al-Sarem et al [ 44 ], Al-rimy et al [ 48 ], Kaushik et al [ 57 ]) compared to several machine learning classifiers. …”

“…al [48], Kaushik et. al [57]) using a set of 10 lected features. Comparisons between the accuracy of the proposed LRGU-MIFS and related features lection techniques (Al-Sarem et.…”

“…al [48], Kaushik et. al [57]) using a set of 10 lected features. The comparison results showed that the proposed LRGU-MIFS technique outperformed the techniques used by the related works.…”