Enhancing Machine Learning Models for Attack Detection Using SMOTE

In cybersecurity, intrusion detection systems are crucial for determining and mitigating attacks. However, the class imbalance in datasets like NSL-KDD, where attack instances are under-represented, can hinder the performance of Mchine Learning models. In this paper, the class imbalance issue is addressed in the NSL-KDD dataset and to overcome this issue and the Synthetic Minority Over-sampling Technique (SMOTE) technique is applied. The evaluation is performed with various machine learning models, including Random Forest, SVM, KNN, Logistic Regression, Naive Bayes, and XGBoost, both before and after applying SMOTE. The results reveal significant improvements in the detection rates of minority attack classes, with reduced False Positive Rates. XGBoost and Random Forest emerged as the best-performing models, achieving accuracies of 95.5% and 93%, respectively, after applying SMOTE. These outcomes show the significance of SMOTE in improving model robustness and ensuring a more reliable intrusion detection system.