| Title |
Analyzing Park Accessibility Mismatch Between Resident and De Facto Populations Using Machine Learning and Explainable AI |
| Authors |
이혜준(Lee, Hye-Jun) ; 박훈태(Park, Hoon-Tae) ; 진재특(Chin, Jae-Teuk) |
| DOI |
https://doi.org/10.38195/judik.2026.08.27.5.27 |
| Keywords |
인공지능; XGBoost; H3 격자; 보행 네트워크; G2SFCA Artificial Intelligence; XGBoost; H3 Grid; Pedestrian Network; G2SFCA |
| Abstract |
This study incorporates de facto population demand to compare park accessibility and equity in Seoul based on resident and de facto populations. It also examines factors associated with accessibility mismatch between the two population measures. Accessibility percentile-rank differences between the two population measures were modeled using XGBoost, with explainable AI examining relative variable importance and nonlinear relationships. The proportion of cells without park access decreased from 48% to 9% as walking threshold increased from 300 to 600m. The residents’ Gini coefficient was higher than the mean across de facto population groups. XGBoost outperformed the linear benchmark, and officially assessed land prices, office floor area, and detached and multi-family housing floor area were key associated factors. Higher land prices and greater office floor area were nonlinearly associated with lower de facto-based accessibility ranks relative to resident-based ranks, whereas greater detached and multi-family housing floor area was associated with higher relative ranks. This study demonstrates the usefulness of explainable AI for interpreting relationships between accessibility mismatch across population measures and urban spatial characteristics. |