Hybrid Feature Selection and Hyperparameter-Optimized Soft Voting for Network Intrusion Detection

Abstract

A great deal of effort has been put into the development and implementation of Network Intrusion Detection Systems (NIDS) for the security of ICT infrastructures against the contemporary cyber-attacks has been one of the basic strategies used. When the traffic being analyzed has high dimensions, however, there might be redundancies and irrelevancies among features which will make the computation process of network traffic more challenging and will reduce the intrusion detection effectiveness. To solve these issues, in this paper, an integrated approach to feature selection and ensemble learning techniques for network intrusion detection problem based on the UNSW-NB15 dataset is proposed. To improve data quality and balance the training data, data cleansing, feature encoding, normalization, and the application of Synthetic Minority Oversampling Technique (SMOTE) are used as pre-processing steps. After that, two hybrid feature selection methods: Mutual Information (MI)-Particle Swarm Optimization (PSO) and Analysis of Variance (ANOVA)-Boruta are used to determine the significant features from the datasets. Selected feature subset is then trained using Random Forest (RF), AdaBoost and XGBoost. The hyperparameters of the classifiers are optimized using RandomizedSearchCV and the fused classifiers are optimized for the classification performance using Soft Voting Ensemble. The experimental analysis results show that the proposed MI–PSO method can greatly reduce the feature space of the original features from 42 to 9, with a high classification accuracy. The classification performance by proposed MI–PSO based Soft Voting Ensemble was Accuracy of 95.83%, Precision of 96.20%, Recall of 97.74%, F1-Score of 96.96% and ROC-AUC of 99.32%. The framework proposed is an effective compromise on the exchange between feature reduction, calculation efficacy and intrusion detection interpretation, suitable for securing modern network and ICT infrastructures that also enables innovation and secure digital infrastructure.