Comparative Study of Machine Learning Algorithms for DDoS Attack Detection in SDN Networks: A Carbon Emission Analysis with Hyperparameter Optimization Using Bayesian Optimization

The rising energy consumption of artificial intelligence (AI) models has sparked concerns about their environmental impact, particularly in high-computation fields like cybersecurity. As machine learning (ML) models become more complex and resource-intensive, optimizing their energy efficiency and sustainability has become a critical challenge. Bayesian Optimization has emerged as an effective approach for hyperparameter tuning, improving both model performance and energy efficiency. This study explores Tree-structured Parzen Estimators (TPE), a variant of Bayesian Optimization that models hyperparameter distributions using density estimation, to optimize the performance and environmental footprint of three widely used ML algorithms—Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost)—for DDoS attack detection in Software-Defined Networks (SDN). Evaluations on two datasets—Dataset 3 (binary classification) and Dataset 4 (multi-class classification)—analyze accuracy, precision, recall, and F1-score, alongside energy consumption and carbon emissions measured via the CodeCarbon. Results show that RF achieves the highest accuracy across both datasets (99.81%) while reducing carbon emissions by 44.6% after optimization of TPE. XGBoost, while slightly less accurate (99.77%), produces the lowest carbon emissions (0.0006 kg CO₂), demonstrating superior energy efficiency. SVM, despite a 35% reduction in emissions, remains the least efficient in energy consumption and exhibits lowest accuracy. These findings highlight the role of Bayesian Optimization in balancing predictive performance with sustainability. This study contributes by demonstrating a quantitative approach to evaluating the trade-off between accuracy and energy efficiency in ML-based DDoS attack detection in SDN, offering insights into selecting environmentally sustainable models.