Intrusion Detection System Using Stacked Ensemble Learning with Light Gradient Boosting Machine and Decision Tree

The contemporary challenge of increased cyberattacks poses risks to individuals and businesses, impacting the availability, confidentiality, and integrity of sensitive data transmitted over networks. The resulting harm can vary from minor service interruptions to significant financial losses. While conventional security measures like firewalls and antivirus software provide an initial layer of protection, there is a recognized need to develop effective Intrusion Detection Systems (IDS). Current IDS solutions incorporate various classifiers, both individual and ensemble, yet challenges persist in identifying novel intrusions accurately. Thiswork introduces a novel model, the Intrusion Detection System, using the Stacked Ensemble Learning Technique (IDSELSE), which addresses these challenges. IDSELSE leverages Kmeans SMOTE oversampling to handle class imbalance and enhance minority class representation. Additionally, it employs Boruta for feature selection, streamlining the model's efficiency by eliminating irrelevant features. In classification, IDSELSE utilizes the Stacked Ensemble Learning Technique with a combination of Light Gradient Boosting Machine (LGBM) and Decision Tree (DT) as base classifiers, supported by the meta-model Logistic Regression (LR), to make accurate predictions. Performance evaluation involves assessing F1-score and accuracy through tenfold cross-validation, demonstrating the consistent superiority of the proposed model over various single-classifier and ensemble models documented in the literature.