A Machine Learning Approach for Generating and Evaluating Forecasts on the Environmental Impact of the Buildings Sector

Abstract

The building sector has traditionally accounted for about 40% of global energy-related carbon dioxide (CO2) emissions, as compared to other end-use sectors. Due to this fact, as part of the global effort towards decarbonization, significant resources have been placed on the development of technologies, such as active buildings, in an attempt to achieve reductions in the respective CO2 emissions. Given the uncertainty around the future level of the corresponding CO2 emissions, this work presents an approach based on machine learning to generate forecasts until the year 2050. Several algorithms, such as linear regression, ARIMA, and shallow and deep neural networks, can be used with this approach. In this context, forecasts are produced for different regions across the world, including Brazil, India, China, South Africa, the United States, Great Britain, the world average, and the European Union. Finally, an extensive sensitivity analysis on hyperparameter values as well as the application of a wide variety of metrics are used for evaluating the algorithmic performance.

Keywords

• ARIMA
• deep learning
• linear regression
• machine learning
• neural networks
• uncertainty