Package: BORG 0.3.2

Installation | Introduction | The Dissimilarity Index | What it measures | Basic usage | Variable importance weighting | Area of Applicability | From DI to a binary mask | Full worked example | Cross-validated threshold | Manual threshold | Extrapolation Detection | Univariate and multivariate checks | Worked example | Comparing extrapolation and AOA | Transferability Assessment | Quantifying performance decay with distance | Interpreting transfer ratios | Prediction Maps | Spatial uncertainty from cross-validation | Stability Maps | Where predictions change with the training set | Stability maps versus prediction maps | Combining AOA with Blocked Cross-Validation | The full pipeline | Diagnose dependencies | Generate blocked CV folds | Fit and evaluate with honest folds | Compute spatially-aware importance | Determine the area of applicability | Generate predictions with AOA mask | Check for extrapolation in the reliable predictions | Practical Guidance | When to use the AOA | When the AOA is less useful | Interpretation guidelines | Mahalanobis distance and its limitations | Reporting AOA results in publications | Connecting to the broader BORG workflow | References

Installation | Introduction | Simulated data | Block-permutation importance | The problem with standard permutation importance | How block-permutation importance works | Worked example | Comparing block vs. row permutation | Forward feature selection | Why greedy search with blocked CV | Step-by-step example | Using a custom model | Interpreting the selection path | Best subset selection | SHAP values | Beyond global importance | Spatial SHAP | When to use SHAP vs. permutation importance | Comparing methods | Full pipeline | Diagnose dependencies | Generate blocked folds | Screen with permutation importance | Forward selection on screened candidates | Fit the final model | Interpret with SHAP | Practical guidance | How many features relative to sample size? | Stability of selection | Collinearity and importance | When NOT to do feature selection | Reporting feature selection in publications | Temporal and grouped data | References

Installation | Why This Guide Exists | Quick Reference | Base R | Safe Preprocessing Pattern | caret | When to Use BORG with caret | Validating preProcess Objects | Guarded trainControl | Pipeline-Level Validation with borg_pipeline() | tidymodels (rsample + recipes) | When to Use BORG with tidymodels | Validating Recipe Objects | Guarded rsample Functions | Converting BORG Folds to rset Objects | Convenience Constructors: borg_spatial_cv() and borg_temporal_cv() | Validating Existing rsample Objects | mlr3 | When to Use BORG with mlr3 | Validating mlr3 Tasks and Resamplings | Generating mlr3 Resamplings from BORG Folds | Species Distribution Models | ENMeval | biomod2 | Temporal Data Workflows | Basic Temporal Validation | Rolling Origin with rsample | Spatial Data Workflows | Spatial Block Validation | Automatic Spatial CV Generation | Grouped Data Workflows | Complete Pipeline Validation | borg_validate(): Workflow-Level Checks | Catching Overlap | borg_pipeline(): Stage-by-Stage Decomposition | Automatic Repair with borg_assimilate() | Migration Checklist | See Also

Installation | Introduction | Simulating spatial data | Quantifying inflation | Fair model ranking | Confidence intervals on metrics | Is the signal real? | Per-fold performance | Full comparison pipeline | Practical guidance | How many repeats for borg_compare_cv() | How many permutations for borg_null_test() | How many bootstrap replicates | When to report which metric | When model comparison is misleading | Reporting for publications | Number of folds | Minimum sample size | References

Installation | Setup: simulating spatial data and fitting a model | Quick diagnostic check | Residual diagnostics | Calibration | Null hypothesis testing | Model comparison | Quantifying inflation with borg_compare_cv() | Ranking models with borg_compare_models() | Conformal prediction | Bootstrap uncertainty | Power analysis | Per-fold performance | Diagnostic workflow | Practical guidance | When to run which diagnostic | Common pitfalls | Integration with external frameworks

Installation | Why your test accuracy might be wrong | Simulating data with known dependencies | Diagnosing data dependencies | Temporal and grouped data | Working with CV folds | Exporting folds to other frameworks | Validating existing splits | Overlapping indices | Target leakage | Preprocessing leakage | Accessing risk details | The borg_check() shortcut | Fitting a model with borg_workflow() | Comparing random vs blocked CV | Power analysis after blocking | Putting it all together | Taste tests: more features | Area of Applicability | Block-permutation variable importance | Residual diagnostics | Calibration | Null test | Extrapolation detection | Geographic distance analysis | Interactive maps | S3 methods | Methods text for publications | Practical guidance | When to use BORG | When NOT to use BORG | Rules of thumb | Validation certificates | Spatial input formats | Function reference | Core | Evaluation | Spatial analysis | Feature selection and importance | Reporting | Simulation | Next steps | References

Installation | Risk Classification | Hard Violations | 1. Index Overlap | 2. Duplicate Rows | 3. Preprocessing Leakage | 4. Target Leakage (Direct) | 5. Group Leakage | 6. Temporal Ordering Violation | 7. CV Fold Contamination | 8. Model Scope | Soft Inflation Risks | 1. Target Leakage (Proxy) | 2. Spatial Proximity | 3. Spatial Overlap | 4. Random CV on Dependent Data | Risk Interactions | Spatial + Target Leakage | Preprocessing + Spatial Structure | Groups + Temporal Ordering | Temporal + Spatial | Decision Tree: Which Risks to Check | Practical Guidance | Hard vs. Soft: Triage Order | How Risk Severity Scales with Dataset Size | Rules of Thumb | Quick Reference | Accessing Risk Details | See Also

Installation | Introduction | The problem: spatial autocorrelation | Simulating spatial data | Spatial block cross-validation | Block size optimization | Alternative spatial CV strategies | Environmental blocking | Leave-disc-out (buffered spatial CV) | Buffered leave-one-out | Spatial+ CV | KNNDM | Distance distribution matching | Comparing random vs blocked CV | Practical guidance | Choosing a spatial CV strategy | Rules of thumb | When NOT to use spatial CV | A practical workflow | References

Installation | Introduction | The problem: temporal autocorrelation | Simulating temporal data | Temporal block CV | Expanding window CV | Embargo periods | Comparing random vs temporal CV | Concept drift detection | Practical guidance | Choosing a strategy | Rules of thumb with specific numbers | When NOT to use temporal CV | Panel data | End-to-end workflow | References