Distributed Intrusion Detection System using Ensemble Learning Technique for Securing IoT Environments

Abstract

Introduction: Conventional anomaly detection systems fall short in Internet of Things (IoT) as the range of possible normal behaviors for devices is much wider and more dynamic compared to traditional environments. Main issues of attack detection in IoT include dataset management and optimal feature selection. But existing Intrusion Detection Systems (IDSs) did not concentrate on this issue.

Objectives: The main objective is to classify the data as normal or anomaly using Machine Learning (ML) based algorithms and to test the dataset against an ensemble model involving number of ML classifiers.

Methods: In this paper, a Distributed Intrusion Detection System (DIDS) for IoT environments, using Ensemble learning technique is designed. Principal Component Analysis (PCA) technique is applied for reducing the dimension of data. For classifying the data as normal or anomaly, the classifiers Decision Tree, Gaussian Naive Bayes and XGBoost are applied. Ensemble Learning technique based on maximum voting is used, which combines the multiple learners in order to improve accuracy.

Results: By experimental results, it has been shown that the proposed maximum voting ensemble technique has higher attack detection accuracy with low false positives. 

Conclusion: This DIDS  can detect and classify nearly 32 attacks..