Predicting CO2 Emission from Power Industry using Machine Learning
Authors
Sankaradass, Harini
Issue Date
2024
Degree
MSc in Data Analytics
Publisher
Dublin Business School
Rights holder
Rights
Abstract
Using a variety of forecasting models, this thesis provides a thorough examination of CO2 emissions from coal-fired electric generating facilities in the United States. The main dataset, which spans the period from January 1973 to August 2023, comes from the U.S. Energy Information Administration. To anticipate CO2 emissions, models such as Prophet, Holt Winter's Exponential Smoothing, and ARIMA were used. After a performance evaluation, ARIMA was shown to be the most efficient model, with the lowest RMSE (2.23), MAE (1.69), and highest R2-score (0.83). Every model was put through a thorough pre-treatment procedure that included exploratory data analysis, feature extraction, and data cleaning. The study
examines these model’s advantages and disadvantages extensively. Notably, Holt Winter's Exponential Smoothing and Prophet showed benefits in managing trend patterns and capturing seasonality, even though ARIMA showed better predicted accuracy overall. The use of these models and its consequences for environmental policymaking are ethical issues that are also covered in the thesis. This study identifies gaps in the literature, especially regarding the possibility for hybrid modelling techniques, ethical frameworks, scalability across different contexts, and the inclusion of external elements. To contribute to more reliable and responsible forecasting frameworks for CO2 emissions from coal-fired power plants, the findings indicate directions for future study aimed at improving forecasting accuracy and ethical applicability.