Tools
Tools
Manivannan, Lakshmi Meena (2025) Green AI: Quantifying the Carbon Footprint of Machine Learning Models Through Multi-Model Energy Benchmarking and Emissions Analysis. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (1MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (1MB) | Preview |
Abstract
The project is an environmental assessment of classical machine learning models, which is done through a combined analysis of their predictive performance, energy consumption and approximate carbon emissions. It uses two real-world datasets one of the air-quality sensor datasets and one of the electricity-consumption time-series datasets. The trained models are Logistic Regression, Support Vector machine, Random Forest, Multilayer Perceptron and XGBoost, which are run in a controlled pipeline and all experiments tracked with the help of CodeCarbon library. The models have a predictive accuracy of 72-82% (R 2 on regression tasks or the same accuracy in classification tasks) the highest scores being achieved by Random Forest and XGBoost. Nevertheless, such sophisticated models produce up to 45x more CO 2 than lighter baselines like Logistic Regression, and do not increase accuracy by more than 2 -3 points. The visualizations of Streamlit dashboard are created to visualise the per-model emissions, runtime and efficiency rankings, and allow a transparent comparison of accuracy-emissions trade-offs, and aid the choice of greener models.
Actions (login required)
 |
View Item |