Modeling public parking demand at the small-block level using detailed building-use data: A market-segmentation approach

Public parking demand in Korean cities has been inadequately modeled because prior studies rely on administrative-district aggregation and zoning- or floor-area-based land-use variables decoupled from actual parking-generating activity. This study introduces 165 parcel-level building-use codes from the Korean building-property tax ledger, aggregated to 137 small blocks in Suseong-gu, Daegu, for the years 2020 and 2023. Factor analysis with Ward's hierarchical clustering identifies three interpretable clusters, and market-segmentation regressions are estimated separately for daytime and nighttime demand. Segmentation improves all reproducibility metrics (MAE, RMSE, MAPE, χ², r) relative to the pooled model across all (year × time) combinations, and a χ² homogeneity test rejects model equivalence at p < 0.001. Medium-density residential floor area is the strongest predictor of nighttime demand in both aggregate and segmentation models. Food-and-wedding floor area leads daytime demand in the aggregate model, while medical and commercial uses are significant daytime predictors within cluster models. Proximity to urban-rail stations is positively associated with daytime demand in the mixed-residential cluster, consistent with a co-agglomeration mechanism obscured by the aggregate model. The findings support time-of-day-differentiated policy: nighttime capacity expansion near medium-density residential zones and daytime expansion near activity-intensive uses. The pipeline is reproducible wherever comparable building-use records exist.