A Comparison of Ensemble Learning for Intrusion Detection in Telemetry Data

The Internet of Things (IoT) is a grid of interconnected pre-programmed electronic devices to provide intelligent services for daily life tasks. However, the security of such networks is a considerable obstacle to successful implementation. Therefore, developing intelligent security systems for IoT is the need of the hour. This study investigates the performances of different Ensemble Learning (EL) approaches applied for intrusion detection in the IoT sensors’ telemetry data. We compare the accuracy of various EL approaches in homogeneous and heterogeneous combinations using bagging, boosting, and stacking strategies. These EL approaches apply well-known Machine Learning (ML) models such as Decision Tree (DT), Naıve Bayes (NB), Random Forest (RF), Logistic Regression (LR), Linear Discriminant Analysis (LDA) and linear Support Vector Machine (SVM). We evaluate and compare EL approaches for binary and multi-class classification tasks on the ToN-IoT Telemetry dataset for intrusion detection. The results show that stacking EL outperform stand-alone ML algorithms-based classifiers as well as bagging and boosting.