Intrusion Detection System Traffic Classification Based on Machine Learning with Correlation-Based Filtering and a Genetic Algorithm-Inspired Feature Selection Method for IoT Networks

Abstract

Securing Internet of Things (IoT) networks against diverse cyber-attacks remains a critical challenge due to their constrained resources and complex traffic patterns. This paper proposes a lightweight, multiclass Intrusion Detection System (IDS) that addresses the limitations of prior binary models by classifying five types of network traffic: Normal, DoS, Mirai, Man-in-the-Middle (MITM), and Scan. A key contribution of this work is the application of a Genetic Algorithm (GA)-inspired feature selection method, which significantly enhances model accuracy and efficiency by isolating the most relevant attributes. Combined with traditional machine learning models, the proposed approach was evaluated using a simulated dataset modeled after IoTID20. Among the classifiers, the Random Forest model, when integrated with GA-inspired feature selection, achieved the highest accuracy of 96.5%. The results highlight the effectiveness of combining lightweight feature optimization with robust classification techniques, making the system highly suitable for real-world IoT deployments.