A Multiclass Network Intrusion Detection System Using Stacking Ensemble Technique with Hybrid Feature Selection

Intrusion detection in network is essential in order to preserve the confidentiality, integrity and authentication in online community.Various research in earlier have used conventional machine learning classifiers and deep learning classifiers to detect intrusion in network but there is a need to achieve improved efficiency of classifiers.In this study we have proposed stacking ensemble machine learning model detect multiclass intrusion in network more effectively.Decision Tree, Random Forest (RF), XGBoost (XGB), AdaBoost, LightGBM (LGBM), Logistic Regression (LR) and Multilayer Perceptron (MLP) were used in ensemble learning for classification of intrusion.The novelty of the work lies in utilizing three techniques for feature selection, namely Gini index, correlation analysis, and random forest to select the most important features.To address class imbalance, Synthetic Minority Over-sampling Technique (SMOTE) was applied.In order to test the effectiveness of the proposed model it was evaluated on UNSWB-NB15, CSE-CIC-IDS2018 and HIKARI-2021 latest heterogeneous datasets.Our proposed method produces excellent results in terms of accuracy and precision, which exceeds 96% and 98%.For better validation other metrics such as Recall, F1-Score, AUC-ROC score, Log loss value and Cohen's Kappa score was considered.From the research findings it can be indicated that the proposed model significantly enhances the multiclass network intrusion detection through its outperforming performance and has potential to strengthen the networks and systems from cyber-attacks.