Comparison of Supervised Machine Learning Algorithms with Bagged-Ensemble Method for Intrusion Detection

Intrusion Detection Systems (IDS) are key parts of cybersecurity frameworks. The primary objective of IDS is to detect and mitigate unauthorized activities or breaches of regulations occurring within a network or system. Machine learning (ML) models are often employed in IDS because of its efficiency in handling the network data and improving detection precision. This work develops an IDS by combining bagged ensemble with ML models. Bootstrap Aggregating, commonly referred to as bagging, is a technique in ensemble learning which combines the predictions generated by numerous base models in order to improve the effectiveness of the system, as opposed to relying on the prediction of a single ML model. Here feature selection techniques are also used for selecting important features instead of all features. The classifiers that are used as base model here are: Logistic Regression (LR), K-nearest neighbor (KNN), Naive Bayes (NB) and Support Vector Machine (SVM). Performance assessments are carried out using the KDDCUP 99 dataset to determine the model that achieves the highest intrusion detection rates. The model's performance is analyzed through the evaluation metrics like F1score, AUC score, accuracy, recall, and precision. Results from the bagged model are compared with the Random Forest classifier. Random forest algorithms have highest accuracy of 99%. The simulated result also shows that proposed bagged ensemble model performs well than traditional model except for Logistic Regression.