Ensemble Technique to Detect Intrusion in a Network Based on the UNSWB-NB15 Dataset

A crucial component of network security is intrusion detection, which guards against attacks and unwanted access to computer systems. Due to their reliance on signature-based detection, traditional intrusion detection systems (IDS) cannot discover unknown advanced threats. Machine learning-based techniques have demonstrated positive results in recognizing unidentified malicious attacks. However, no learning algorithm-based model can reliably and precisely identify every type of attack. Aside from that, a particular dataset is used to test the current model. This study is carried out as a preliminary work for network intrusion detection as the model is tested with a single dataset for binary classification, where the study can be further extended to imbalanced datasets to test the model's robustness. The work has employed machine learning techniques for intrusion detection based on the stacking ensemble approach. The model's performance is evaluated for binary classification with standard metrics on UNSWB-NB15 dataset. The stacking ensemble technique is constructed with a Decision Tree, Random Forest, K-Nearest Neighbor, XGBoost, Logistic Regression, and Multilayer Perceptron. The preliminary results have shown that the ensemble method has outperformed the performance of the standalone model. Empirically, the proposed method outperforms the works studied in the literature by a significant margin for binary classification. This proposed method can be a useful defense technique to safeguard the network and its resources from the hands of cyber-attacks.