Skip to main content
Search

# Histogram-based Gradient Boosting Regression

1 related articles

Subscribe to updates
  1. Leakage-aware ML predicts MOF drug loading and cell viability at in-domain R² of 0.557 and 0.774, but R² turns negative when whole publications are held out

    Using data reconstructed from Wang et al.'s supplementary tables, this study built 161 loading-capacity observations with 110 descriptors and 444 cell-viability observations with 25 descriptors, optimized histogram-based gradient boosting regression (HGBR) and partial least squares regression (PLSR) with differential evolution under five-fold grouped cross-validation, and constrained exact duplicate records to the same partition to prevent information leakage; on the duplicate-safe 20% holdout HGBR was strongest for cell viability (R² = 0.774, RMSE = 11.655 percentage points, MAE = 8.153, AARD = 16.69%) while PLSR was best for loading capacity (R² = 0.557, RMSE = 0.345 g/g, MAE = 0.210 g/g), yet holding out entire source publications drove all R² values negative (viability −0.391 and −0.

You’re all caught up