Entropy-Guided Jaya optimization for robust feature selection in high-dimensional network attack detection

High-dimensional network traffic data pose a significant challenge for the design of efficient and accurate Network Attack Detection Systems (NADS). Choosing features is essential to minimize redundancy, improve generalization, and facilitate real-time analysis. In this work, we propose an Entropy-Guided Jaya (EGJ) optimization framework for feature selection in DDoS detection tasks. The framework employs a weighted composite entropy measure, combining Shannon entropy and conditional entropy, to initialize the Jaya population with informative feature subsets. This initialization is followed by iterative optimization using a binarized and reinforced variant of the Jaya algorithm, which balances exploration and exploitation in the high-dimensional search space. To evaluate the discriminative power of the selected features, we employ the XGBoost classifier implemented on the H2O platform, chosen for its scalability and robustness on large-scale security datasets. Experimental evaluation on the CIC-DDoS2019 benchmark demonstrates that the proposed method achieves substantial feature reduction while maintaining high recall, precision, and F1-score. It attains an average accuracy of 93%. The results confirm that entropy-guided feature selection, combined with ensemble-based classification, offers a lightweight yet effective approach to secure modern networks against DDoS attacks.