A Prediction Model for Greenhouse Gas Emissions and Patterns across Boundary Scopes in the Construction Phase – Utilizing Explainable Machine Learning

The recent intensification of global warming, often referred to as "global boiling" due to climate change, has emerged as a critical societal challenge, accelerating global efforts to achieve carbon neutrality through greenhouse gas (GHG) reduction initiatives. South Korea has set a target to achieve carbon neutrality by 2050 and aims to reduce GHG emissions by 40% compared to 2018 levels by 2030. Consequently, the construction industry is also under increasing pressure to develop sustainable and effective strategies for reducing GHG emissions. However, existing GHG management systems in construction often fail to adequately account for project-specific characteristics, limiting their capability to accurately predict and manage the total volume and patterns of emissions. This study proposes a model capable of precisely predicting GHG emissions and their patterns across boundary scopes (Scopes 1,2, and 3) during the construction phase. The proposed model utilizes a Case-Based Reasoning (CBR) methodology, incorporating a filtering engine based on multiple machine learning models to achieve both explainability and high predictive accuracy. The average prediction accuracy (APA) of the proposed model demonstrated excellent performance, with Scope 1 at 80.2%, Scope 2 at 92.8%, and Scope 3 at 80.0%. The proposed model supports government agencies in establishing rational emission allowances tailored to project characteristics and assists construction firms in identifying GHG emission risks at an early stage, enabling the development of effective mitigation strategies. The proposed model is anticipated to become a practical tool for GHG management in the construction sector, making a significant contribution to the industry's carbon neutrality objectives.