A Robust Intrusion Detection System with Randomized Search and Balanced Ensemble Models

Resumen

Intrusion detection systems are crucial for protecting the network infrastructure from attacking activities. The effectiveness of these systems may be hampered by class imbalance, feature redundancy, and high-dimensional datasets. In the study, we introduce a robust anomaly detection system based on the Synthetic Minority Over-sampling Technique (SMOTE), feature engineering, and ensemble learning. The study uses Support Vector Machines, Logistic Regression, Decision Trees, and Random Forest for the Voting Classifier framework powered by a Randomized Search Cross-Validation. This approach underwent feature engineering that resulted in a reduction in the dimensionality and limitation to multicollinearity, while SMOTE was concerned with the balance of classes. The model provided achieved an impressive accuracy of 99.72%, along with macro average scores of 0.95 for precision, 0.85 for recall, and 0.88 for the F1 score. The classification report showed that the classifier worked perfectly for the majority classes and also maintained relatively good performance for the minority classes, raising the issue of its superior performance. The proposed work shows that advanced resampling methods and ensemble learning stand as strong tools for intrusion detection in complicated network settings.