An Efficient IDS For Iot Security Using Ensemble Machine Learning

Abstract

Internet-of-Things (IoT) is gaining popularity because of its ability to interconnect a variety of devices which can exchange data with ease. However, its pervasiveness makes it vulnerable to security threats posed by intruders. Therefore, building robust Intrusion Detection Systems (IDS) to protect IoT components from intruders is a challenging task. In this paper, an attempt has been made to build anomaly-based network intrusion detection model using ensemble methods namely, AdaboostM1, bagging, random forest, and dagging. Further, in order to enhance the performance of the model, three feature selection methods namely, chi-square, symmetrical uncertainty, and correlation attribute have been applied to remove the non-contributing features from the dataset. Also, data normalization has been applied on the selected features for scaling. The RT IoT 2022 dataset has been used for model building and evaluation. Comparative analysis shows that though all models performed well but the model Relief F + XGBoost emerges as the best-performing model, achieving the highest accuracy of 0.9990, precision of 0.9995, specificity of 0.9955, KC of 0.9945, GM of 0.9974, and YI of 0.9949, alongside the lowest FPR of 0.0045, RAE of 0.0013, MAE of 0.0010, and RMSE of 0.0317, and an F-value of 0.9994 tied for the highest in the table. Although Gain Ratio + Bagging reports a marginally higher MCC of 0.9985 compared to 0.9945 for Relief F + XGBoost.