Strategic Data Augmentation for IoT Intrusion Detection Using Ensemble Machine Learning

Abstract

The Internet of Things (IoT) and modern web-based infrastructure are increasing at a very fast rate, resulting in a rise in complex and compact cyber-attack patterns that tend to evade traditional signature-based protection methods. A key trade-off in the detection capability and rate of false alarms arises from the prevailing methods' common difficulty in addressing the “Class Imbalance” problem that exists in realistic network traffics, although Machine Learning has shown promise as a potential paradigm for Network Intrusion Detection Systems (NIDS). The “Zero False Negative” challenge in high-security applications remains unaddressed in prevailing state-of-the-art methods that tend to plateau at very high levels of accuracy (around 99.6 percent). This paper proposes a novel High-Redundancy Network Intrusion Detection Framework that relies on a Strategic Data Augmentation approach that can address the “Zero False Negative” challenge.

The proposed framework tests three distinct classifiers, namely Random Forest, Extreme Gradient Boosting (XGBoost), and a constrained Decision Tree (for baseline complexity testing). Based on the experimental results, the Random Forest and XGBoost models were able to fill the performance gap created in the previous study by achieving 100.00% Accuracy, 1.00 Precision, and 0.00% False Alarm Rate (FAR) due to the proposed augmentation method. However, the baseline Decision Tree could only achieve 89.09% accuracy, proving that the dataset still retains a large amount of structural complexity that demands robust ensemble learning. Although both ensemble methods made perfect detections, XGBoost is more efficient in computation and converged 3.3 times faster (21.48 seconds) than Random Forest in an efficiency comparison study. Hence, in real-time Zero Trust Network Intrusion Detection in high velocity networks, the proposed study recommends the Augmented XGBoost model to be the most optimal choice.