Online ML Streaming-Based Leading Ensemble Classifier for Intrusion Detection in IoT

Abstract

In the present decade, preventing IoT networks from malicious attacks and to ensure the privacy and integrity of data by strengthening their cybersecurity perspective is a critical concern. Traditional methods alone are not sufficient to detect multiple attacks. IDS-based system plays a crucial role to maintain security due to continuous expanding nature of IoT networks. Traditional methods alone are often insufficient to detect and counter attacks that try to impact IoT devices. This creates a demand for advanced technological approaches to boost the security of IoT networks. One approach gaining attraction in addressing this challenge is by deploying ensemble machine learning models for identifying the attack categories and preventing them. To build an ensemble framework for the identification of malicious behavior in IoT-based network, this study combines the results and metrics obtained from 4 base classifiers LightGBM, Random forest classifier (RFC), XGBoost, and CatBoost. Merged them into an online ML streaming-based Leading Ensemble Decision Classifier Module (LEDCM). This module predicts the accurate class for every record in both multiclass and binary classification scenarios using IoT-based datasets (i.e. NSL-KDD, NF-BOT-IoT, UNSW-NB-15 datasets), LEDCM merge the prediction from all the major 4 classifiers in such a way, the overall Accuracy % and f1-score will get increased by fetching the best prediction from all base classifiers in a group individually Class wise. In the final results, we observed that our proposed model achieved the highest accuracy of 99.94% for NSL-KDD test data, 95.99% for UNSW-NB-15, and 83.13% for NF-BOT-IoT for Multiclass classification, and Accuracy of 100% for all three Datasets in the case of binary label classification in comparison to other base classifiers.