Assessing the Performance of Spatial Cross-Validation Approaches for Models of Spatially Structured Data

article
software
machine learning

Mahoney MJ, Johnson LK, Silge J, Frick H, Kuhn M, Beier CM (2023). “Assessing the performance of spatial cross-validation approaches for models of spatially structured data.” arXiv preprint arXiv:2303.07334.

Abstract

Model stacking is an ensemble modeling technique that involves training a model to combine the outputs of many constituent statistical models. {stacks} is a free and open-source R software package for stacked ensemble modeling that is consistent with tidy data principles. The package’s functionality is closely aligned with the {tidymodels}, a collection of packages providing a unified interface to a diverse set of statistical modeling techniques. Beyond simply providing a mathematically robust interface to build stacked ensemble models, {stacks} adheres to a consistent grammar in order to interface with two object classes that promote an intuitive understanding of the underlying implementation.