<![CDATA[The increasing number of network attacks requires intrusion detection models that are not only accurate but also efficient in terms of execution time, energy use, and carbon emissions. This study evaluates baseline XGBoost and four hyperparameter optimization approaches, namely Tree-structured Parzen Estimator (TPE), Asynchronous Successive Halving Algorithm (ASHA), Hyperband, and Particle Swarm Optimization (PSO), for multi-class intrusion classification on the CICIDS2017 Wednesday-workingHours subset. The evaluation uses predictive metrics, including accuracy, macro recall, macro F1-score, macro precision, and one-vs-rest AUC, as well as computational sustainability metrics consisting of CO2 emissions, energy consumption, execution time, and CPU/RAM/GPU energy details. The results show that XGBoost+TPE achieves the highest predictive performance with an accuracy of 0.9995 and macro F1-score of 0.9959, although its execution time increases to 13.04 seconds. In contrast, XGBoost+Hyperband maintains an accuracy of 0.9994 and macro F1-score of 0.9957 while reducing CO2 emissions by 56.50%, energy consumption by 56.54%, and execution time by 49.60% compared with the baseline. These findings indicate that Hyperband provides the most balanced configuration for intrusion detection scenarios that require high accuracy and computational efficiency.]]>
Copyrights © 2026
