Energy-Efficient Machine Learning Algorithms for Sustainable AI Systems

Abstract

The rapid expansion of Artificial Intelligence (AI) systems across enterprise and industrial domains has led to a substantial increase in energy consumption, raising concerns about sustainability and operational efficiency. This research presents a comprehensive method for energy-efficient machine learning (ML) algorithms to develop sustainable AI systems while maintaining high predictive performance. Min-Max normalization normalizes feature ranges to improve model stability and efficiency, while Principal Component Analysis (PCA) reduces dimensionality, removing redundancy to enhance computational efficiency and energy-aware learning aligned with sustainable AI objectives. ML models, including Dynamic Raven Roosting Optimized Enriched Support Vector Machines (DRRO-En-SVM) for robust classification and complex pattern recognition, are integrated to address diverse analytical tasks while optimizing energy usage. The primary purpose of these ML techniques is to enhance prediction accuracy, automate intelligent decision-making, and reduce computational overhead through optimized model design. The proposed method incorporates energy-aware strategies such as model pruning, quantization, and adaptive learning mechanisms to minimize power consumption during both training and inference stages. An optimization approach is employed to balance energy efficiency and model accuracy, identifying Pareto-optimal solutions for different deployment scenarios. Experimental evaluation on large-scale datasets demonstrates that the DRRO-En-SVM (Proposed) achieved 98.9% accuracy, 3.1% error rate, 98.5% F1 score, 98.7% precision, 98.2% recall, and 1.05 seconds of training time. The results highlight the effectiveness of integrating energy-efficient ML techniques in enabling scalable, cost-effective, and environmentally sustainable AI systems, and providing insights for future advancements in green AI technologies.