BORG
Comprehensive toolkit for valid spatial, temporal, and grouped model evaluation. Automatically detects data dependencies (spatial autocorrelation, temporal structure, clustered observations), generates appropriate cross-validation schemes (spatial blocking, checkerboard, hexagonal, KNNDM, environmental blocking, leave-location-out, purged CV), and validates evaluation pipelines for leakage. Includes area of applicability (AOA) assessment following Meyer & Pebesma (2021) <doi:10.1111/2041-210X.13650>, forward feature selection with blocked CV, spatial thinning, block-permutation variable importance, extrapolation detection, and interactive visualizations. Integrates with 'tidymodels', 'caret', 'mlr3', 'ENMeval', and 'biomod2'. Based on evaluation principles described in Roberts et al. (2017) <doi:10.1111/ecog.02881>, Kaufman et al. (2012) <doi:10.1145/2382577.2382579>, Kapoor & Narayanan (2023) <doi:10.1016/j.patter.2023.100804>, and Linnenbrink et al. (2024) <doi:10.5194/gmd-17-5897-2024>.
- Cross‐validation strategies for data with temporal, spatial, hierarchical, or phylogenetic structure
- kNNDM CV: k -fold nearest-neighbour distance matching cross-validation for map accuracy estimation
- Leakage and the reproducibility crisis in machine-learning-based science
- Leakage in data mining
- Predicting into unknown space? Estimating the area of applicability of spatial prediction models
This page was built for software: BORG
