Enhancing Intrusion Detection Systems with XGBoost Feature Selection and Deep Learning Approaches

Abstract

As cyber-attacks evolve in complexity and fre-quency; the development of effective network intrusion detection systems (NIDS) has become increasingly important. This paper investigates the efficacy of the XGBoost algorithm for feature selection combined with deep learning (DL) techniques, such as ANN, 1DCNN, and BiLSTM, to create accurate intrusion detection systems (IDSs) and evaluating it against NSL-KDD, CIC-IDS2017, and UNSW-NB15 datasets. The high accuracy and low error rate of the classification models demonstrate the potential of the proposed approach in IDS design. The study applied the XGBoost feature extraction technique to obtain a reduced feature vector and addressed data imbalance using the synthetic minority oversampling technique (SMOTE), signif-icantly improving the models’ performance in terms of precision and recall for individual attack classes. The ANN + BiLSTM model combined with SMOTE consistently out performed other models within this paper, emphasizing the importance of data balancing techniques and the effectiveness of integrating XGBoost and DL approaches for accurate IDSs. Future research can focus on implementing novel sampling techniques explicitly designed for IDSs to enhance minority class representation in public datasets during training.