Intrusion Detection in NSL-KDD Dataset Using Hybrid Self-Organizing Map Model

Abstract

Intrusion attempts against Internet of Things (IoT) devices have significantly increased in the last few years. These devices are now easy targets for hackers because of their built-in security flaws. Combining a Self-Organizing Map (SOM) hybrid anomaly detection system for dimensionality reduction with the inherited nature of clustering and Extreme Gradient Boosting (XGBoost) for multi-class classification can improve network traffic intrusion detection. The proposed model is evaluated on the NSL-KDD dataset. The hybrid approach outperforms the baseline line models, Multilayer perceptron model, and SOM-KNN (k-nearest neighbors) model in precision, recall, and F1-score, highlighting the proposed approach’s scalability, potential, adaptability, and real-world applicability. Therefore, this paper proposes a highly efficient deployment strategy for resource-constrained network edges. The results reveal that Precision, Recall, and F1-scores rise 10%–30% for the benign, probing, and Denial of Service (DoS) classes. In particular, the DoS, probe, and benign classes improved their F1-scores by 7.91%, 32.62%, and 12.45%, respectively.