Efficient Hybrid Intrusion Detection Approach based on BPR-GWO for Network Traffic Classification and Improved Network Security
International journal of intelligent engineering and systems · 2025
As network traffic complexity and volume grow, the design of accurate, efficient intrusion detection systems (IDSs) becomes an urgent concern.Classic classifiers such as Naive Bayes (NB), Logistic Regression (LR), and Random Forest (RF) have been frequently employed in IDS; however, reliability issues often arise due to insufficient representation of uncertainty along with limited abilities to depict nonlinear data distributions.These drawbacks cause reduced accuracy and increase false detection rates that are very prominent in the current network environment.In all the challenges mentioned earlier, we propose a novel classification framework for network traffic classification based on Bayesian Probit Regression (BPR).The BPR model uses the probabilistic approach in classification, which, compared with conventional classifiers, uses the underlying distribution of traffic behaviour to distinguish normal patterns from malicious ones more accurately.The proposed method was implemented using the Grey Wolf Optimizer (GWO) to enhance detection performance while minimizing incidental overhead from feature selection.GWO thereby reduced the features from 41 to 17 for the NSL-KDD dataset and from 49 to 20 for the UNSW-NB15 dataset.The integrated system with the BPR classifier achieved accuracies of 99.20% and 99.88% on NSL-KDD and UNSW-NB15, respectively.The average detection time per network traffic sample for both was 4.3 ms and 3.7 ms in terms of efficiency.Hence, not only GWO-BPR worked better in the aspects of detection accuracies compared to traditional classifiers, but it also proved to be more efficient; hence, it can be efficiently deployed, even in intelligent real-time IDS solutions.