Hybrid Particle Swarm Optimization and Recursive Feature Elimination for Enhanced Network Intrusion Detection

With the rapid development network technologies security has become more analytical problem. Now a days, Network Intrusion detection System (IDS) place a vital role for reliable and secure data communication. It is a software that detects anomalies or attack pockets and notifies the administrator. Several studies have been proposed by numerous scholars for strengthening IDS performance. However, there are certain challenges for identifying novel attacks and anomalies. Our Studies recommended to improve the IDS performance and identifying novel attacks and reduce the false alarm rate. In this context, initially analyzed well known NSL KDD dataset using traditional classifiers such as Logistic Regression (LR). Decision Tree (DR), Random Forest (RF), Gradient Boosting Tree (GBT), Support Vector Machine (SVM) and Naïve Bayes (NB). Based on this analysis, we have obtained 86.69% test accuracy from random forest classifier using the NSL KDD test set due to imbalanced dataset, outliers and bias and variance issues of this dataset. To improve the model performance, outliers are identified and removed using Isolation Forest (IF). Particle Swarm Optimization (PSO) is used to select best features, yielding a subset of highly informative attributes and reduce model running time. Synthetic Minority Oversampling Technique (SMOTE) is used to handle imbalanced issues, followed by Recursive Feature Elimination (RFE) is used to further reduce the feature set while preserving predictive power. The enhanced Random Forest classifier achieves high detection accuracy of 86.99%, according to experimental results. The proposed pipeline improves model generalization, minimizes imbalance-related bias, and efficiently reduces dimensionality.