Ensemble-Based Lightweight Machine Learning Optimization for IoT Network Intrusion Detection

The rapid development of the Internet of Things (IoT) presents various security concerns, necessitating the implementation of robust and effective Intrusion Detection Systems (IDS). With the growth of IoT networks, cyber risks have increased, putting data integrity and stability at risk. As consequently, the growth of IoT networks has coincided with an increase in cyber threats, threatening the security and integrity of data. Various IDSs have been suggested to identify malicious activity based on previously established attack patterns; this should help to protect data from abuse and odd attempts. Existing intrusion detection systems need upgrading due to the fast rise in these types of threats. This study proposes an Ensemble-Based Lightweight Machine Learning (ELML) specifically optimized for IoT network IDS, leveraging the UNR-IDD dataset. The UNR-IDD dataset is a recent and extensive benchmark that includes several contemporary network attack scenarios, facilitating effective training and assessment of detection algorithms. The proposed approach utilizes feature engineering and lightweight machine learning models including K-Nearest Neighbors (KNN), Random Forest (RF), Logistic Regression (LR), AdaBoost (AB), and Decision Tree (DT). In addition, the Ensemble-Based Lightweight Machine Learning (ELML) IDS model is being optimized for performance using a hyperparameter optimization technique that is based on the Crow Search Algorithm (CSA). The results show significant accuracy improvements over existing algorithms in multi-class classification contexts.