A Comparative Examination of Bagging Techniques for Intrusion Detection System

This study presents a comparative analysis of Bagging and Boosting ensemble techniques for Intrusion Detection Systems (IDS) using the UNSW-NB15 dataset. The dataset, containing both real-world and as well as artificial attack scenarios, serves as a benchmark for evaluating model effectiveness in identifying network intrusions. The study assesses the classification accuracy, stability, and overall performance of various ensemble methods to determine their suitability for IDS applications. Bagging techniques, including Random Forests, k-Nearest Neighbors. (KNN), and Bagged Decision Trees, achieved accuracy rates of 81%, 80%, and 61%, respectively. In contrast, Boosting models, such as Logistic Regression, KNN, and Boosted Decision Trees, recorded accuracy rates of 63%, 79%, and 60%, respectively. The results indicate that Bagging methods, particularly Random Forest, consistently outperform Boosting models in terms of accuracy and stability. While Boosting can enhance learning by focusing on misclassified instances, its performance is less consistent in handling complex attack patterns. In contrast, Bagging effectively reduces variance and enhances model robustness, making it a more reliable choice for IDS. These results emphasize the significance of choosing a suitable ensemble technique depending on dataset traits and performance prerequisites. The study provides insights that can aid in designing more effective IDS solutions for real-world cybersecurity applications.