| Title |
Identifying Key Factors Affecting Apartment Housing Project Schedules With an Integrated AutoML-Cat Boost ?SHAP-PLS Framework |
| Authors |
허영기(Huh, Young-Ki) ; 전정호(Jeon, JungHo) |
| DOI |
https://doi.org/10.5659/JAIK.2025.41.10.359 |
| Keywords |
Apartment Housing; Construction Duration; AutoML; SHAP; PLS |
| Abstract |
Repeated schedule overruns in Korean apartment projects highlight the need for prediction models that are both accurate and interpretable.
This study introduces an AutoML-based CatBoost?SHAP?PLS framework for estimating story-normalized construction duration and identifying
its main drivers. A dataset of 103 apartment projects completed between 2016 and 2019 with 17 independent variables was analyzed. Among
15 candidate algorithms tested using 5x3 repeated cross-validation, CatBoost delivered the best performance, with a root-mean-square error of
5.84 days per story and an R² of 0.38. SHAP analysis revealed that scale-related factors, such as maximum floor and number of households,
reduced duration through repetition efficiency, while subsurface conditions, including basement depth, below-water-table work, and pile count,
increased delays. An exploratory two-component PLS regression explained 46 percent of the variance, showing a negative standardized effect
for height (β??0.51) and a positive effect for basement depth (β?+0.44). The proposed framework integrates nonlinear prediction with
coefficient-level interpretation, enabling contractors and owners to better quantify schedule buffers and prioritize risk controls. |