A Hybrid Feature Selection Model for Anomaly-Based Intrusion Detection in IoT Networks

The ubiquity of the Internet of Things (IoT) network in current trends highlights its integration of billions of interconnected devices and individuals, fostering communication, data sharing, and collaboration. However, this advancement has encountered challenges, with security being the foremost concern. Consequently, users and their data are at risk of cyber-security attacks. The Intrusion Detection System (IDS) plays an essential role in overseeing the network to prevent potential attacks. However, it encounters challenges in IoT networks, where vast data, characterized by a multitude of features for each traffic sample, strains the resources of the IoT network. Dataset imbalance is also considered a common challenge in most IoT datasets. This paper suggested a hybrid handcrafted feature selection technique for efficient IDS for IoT Networks to address these issues. To assess the performance of the proposed hybrid feature selection technique, we employed four machine learning algorithms (Decision Tree, Random Forest, Support Vector Machine, and Gaussian Naive Bayes) on two benchmark datasets: NSL-KDD and BoT- IoT. Our evaluation demonstrates that the Random Forest model achieved outstanding results. It attained an accuracy of 98.97%, 99.9998% and F1-score of 98.97%, 99.53% for both datasets, respectively. Furthermore, the model achieved a high Area Under Curve of 98.98% with low False Positive Rates of 0.82 % and 0.0001 %. Additionally, the model demonstrates an acceptable latency of 0.2 seconds while maintaining a high throughput of over 4 million instances per second, suggesting its potential for efficient and accurate real-time intrusion detection in large IoT networks.