Voting Classifier-based Intrusion Detection for IoT Networks

TL;DR

This paper proposes a voting classifier ensemble method for intrusion detection in IoT networks, achieving high accuracy on sensor data and outperforming traditional machine learning techniques.

Contribution

It introduces a novel ensemble-based voting classifier approach specifically designed for IoT intrusion detection, demonstrating improved accuracy over existing methods.

Findings

Achieved 96% accuracy on GPS sensors

Achieved 97% accuracy on weather sensors

Outperformed traditional machine learning algorithms

Abstract

Internet of Things (IoT) is transforming human lives by paving the way for the management of physical devices on the edge. These interconnected IoT objects share data for remote accessibility and can be vulnerable to open attacks and illegal access. Intrusion detection methods are commonly used for the detection of such kinds of attacks but with these methods, the performance/accuracy is not optimal. This work introduces a novel intrusion detection approach based on an ensemble-based voting classifier that combines multiple traditional classifiers as a base learner and gives the vote to the predictions of the traditional classifier in order to get the final prediction. To test the effectiveness of the proposed approach, experiments are performed on a set of seven different IoT devices and tested for binary attack classification and multi-class attack classification. The results illustrate prominent accuracies on Global Positioning System (GPS) sensors and weather sensors to 96% and 97% and for other machine learning algorithms to 85% and 87%, respectively. Furthermore, comparison with other traditional machine learning methods validates the superiority of the proposed algorithm.