A Green AI Based Approach to Optimizing Machine Learning Models for Sustainable Computing

Abstract

Deep learning has brought notable improvements in predictive performance across a wide range of application areas. At the same time, these advances have led to increased model complexity, higher energy consumption, and a growing carbon footprint. Despite these concerns, much of the existing machine learning research continues to prioritize accuracy, often paying limited attention to energy efficiency and environmental impact. To address this challenge, Green Artificial Intelligence (Green AI) has emerged as an approach that emphasizes the development of energy-aware and environmentally responsible AI systems. In this work, a Green AI–driven framework is presented for optimizing machine learning models with the goal of supporting sustainable computing. The proposed approach combines model optimization techniques—such as pruning, quantization, and knowledge distillation—with lightweight neural network architectures. Energy consumption and carbon emissions are measured during both training and inference using the CodeCarbon framework. Experimental studies are carried out on standard image classification datasets, including MNIST and CIFAR-10, using ResNet, MobileNet, and EfficientNet models. The results show that the optimized models achieve substantial reductions in energy usage and carbon emissions while preserving competitive accuracy. Overall, the proposed framework offers a practical and reproducible solution for deploying energy-efficient machine learning models across both cloud and edge environments.