Enhancing Intrusion Detection Using Hybrid Long Short-Term Memory and XGBoost

Abstract

The application of Long Short Term Memory (LSTM) networks in Intrusion Detection Systems (IDS) is a promising area of research that leverages the strengths of deep learning in sequence modeling and anomaly detection. This paper introduces a proposed enhancement to IDS by designing a hybrid model combining LSTM networks and eXtreme Gradient Boosting (XGBoost). The paper aim is to highlight the limitations of traditional IDS methods, such as low detection accuracy and high false-positive rates, by leveraging the complementary strengths of deep learning and gradient-boosted decision trees. Moreover, the proposed model is designed for multiclass classification, enabling it to accurately detect and differentiate between various types of network intrusions. The proposed approach improves detection accuracy, reduce false-positive rates, and enhance real-time intrusion detection capabilities, thus providing a robust and efficient solution for network security. The results of the experiments show that the proposed model achieves 98.98% accuracy, 99.03% precision, 99.00% recall and 99.02% f1-score on the testing set. The results approve that the proposed model is outperforming most recently proposed models.