A firewall model for attack detection using machine learning and metaheuristic feature selection algorithms

Abstract

This research presents a firewall model designed to enhance network attack detection by integrating machine learning (ML) and advanced feature selection techniques. The study introduces a union-based (DAUBA) feature selection method that combines the exploratory capability of the Dragonfly Algorithm (DA) with the exploitation efficiency of the Bat Algorithm (BA). By combining these two bio-inspired optimizers, the method generates complementary feature subsets that enhance both accuracy and efficiency. The proposed DA?BA feature selection method is incorporated into a ML–based firewall and evaluated on the UNSW-NB15 dataset using three classifiers: adaptive boosting (AdaBoost), K-nearest neighbor (KNN), and Naïve Bayes (NB). Experimental results demonstrate that the approach achieves near-perfect accuracy (100% with AdaBoost), along with strong precision, recall, and F1-scores, while maintaining computational costs compatible with real-time deployment. These findings highlight the novelty and practical value of combining DA and BA in feature selection for next-generation firewall systems.