An Enhanced Intrusion Detection Classification Approach for Securing IoT Networks

Abstract

This study examines multi-class classification techniques commonly used to develop intrusion detection system (IDS). The rapid expansion of internet of things (IoT) has led to an exponential growth in datasets, making it more critical than ever to detect and prevent malicious activities effectively. This research aims to improve IDS performance by using machine learning (ML) algorithms to accurately identify and classify intrusions. The study used three datasets, IoT-Modbus, IoT-Fridge, and IoT-Weather to assess how well classification algorithms can identify and predict various threats in IoT environments. The one-vs-one (OvO) classification approach achieved accuracies of 100% for IoT-Fridge dataset, 99.7% for IoT-Weather, and 77.62 for IoT-Modbus. Similarly, the one-vs-rest (OvR) classification approach achieved accuracies of 100% for the IoT-Fridge dataset, 98.02% for IoT-Weather, and 77.62% for IoT-Modbus. Both approaches performed similarly well on the IoT-Modbus and IoT-Fridge datasets, but the OvO method showed slightly better accuracy on the IoT-Weather dataset. These findings highlight the effectiveness of multi-class classification techniques in IDS and their potential to enhance cybersecurity in IoT applications. This work aims to create a robust IDS that enhances threat detection in the IoT landscape and addresses the unique challenges of IoT security.