Evaluating The Energy Efficiency of Machine Learning For Nids: A Comparative Study Between Xeon Cpu and Rtx Gpu Architectures

The escalating computational demand of machine learning research has historically prioritized state-of-the-art performance at the expense of environmental sustainability, a paradigm increasingly characterized as Red AI. This work addresses the transparency gap regarding the environmental costs of artificial intelligence by providing a comprehensive empirical analysis of Network Intrusion Detection Systems (NIDS) across distinct hardware architectures. Using the CIC-IDS Collection, we evaluate the trade-offs among statistical accuracy, execution time, and carbon emissions in Central Processing Unit (CPU)-based (Intel Xeon Gold) and Graphics Processing Unit (GPU)-accelerated (NVIDIA RTX A5000) environments. Furthermore, we investigate the impact of dimensionality reduction as an energy-optimization lever, contrasting an optimized set of 20 critical features against the original 57 variables. Our results demonstrate that GPU acceleration acts as a fundamental pillar of sustainability. By drastically reducing execution time, it achieves a carbon footprint reduction exceeding 95% compared to CPU baselines. We show that it is feasible to implement robust cybersecurity systems under the Green AI paradigm, maintaining F1-Scores above 93% while significantly minimizing ecological impact. These findings provide actionable guidelines for practitioners to balance high-performance security requirements with environmental responsibility and resource efficiency.