10th International Symposium on Reliability Engineering and Risk Management(ISRERM2026), June 28-July 1, Sapporo, Japan
Environment|구충완
Comparative evaluation of time-series clustering methods for pattern-based classification of carbon emissions in construction projects
2026International Conference
Research AreaEnvironment
Professor구충완
AuthorsAhn, S., Seo, S., Choi, D., Choi, Y., Lee, W., and Koo, C. (Corresponding Author)
Publication Year2026
Journal (Volume, Issue, Pages)10th International Symposium on Reliability Engineering and Risk Management(ISRERM2026), June 28-July 1, Sapporo, Japan
AbstractIn response to Korea’s enhanced 2030 Nationally Determined Contributions (NDC), the construction industry must transition toward more structured and proactive carbon management. Among site-level emissions, Scope 3 emissions account for a substantial share—approximately 52%—yet remain challenging to manage due to their indirect and project-specific characteristics. Rather than relying solely on aggregate emission totals, identifying recurrent time-series emission patterns from historical projects offers a data-driven pathway for early-stage forecasting and benchmarking. To address this challenge, this study aims to establish a methodological foundation for pattern-based classification of Scope 3 emissions through comparative analysis of time-series clustering techniques. First, raw monthly Scope 3 emission data from completed construction projects were structured and normalized to define consistent emission-pattern representations. Second, multiple time-series clustering algorithms were applied and systematically compared to determine their suitability for classifying emission trajectories. Model performance was evaluated using three complementary criteria: (i) cluster balance assessed via the coefficient of variation; (ii) cluster compactness and separation measured by the Davies–Bouldin index and Silhouette score; and (iii) statistical distinctiveness validated through independent t-tests. The results identify clustering approaches that provide stable, interpretable, and statistically differentiated emission-pattern groupings. The proposed framework establishes a reliable historical emission-pattern database that can support future supervised machine learning models for early-stage prediction. By enabling new projects to be benchmarked against historically similar emission trajectories during the planning stage, the approach contributes to more targeted Scope 3 reduction strategies and strengthens carbon management aligned with national NDC objectives.
Scope 3 Carbon EmissionsTime-Series ClusteringEmission Pattern ClassificationEarly-Stage BenchmarkingConstruction Carbon ManagementMachine Learning Foundation