interpretable machine learning

How do the predictions of machine-learning and geostatistical models deteriorate with increasing distance to the training locations? Are different predictors instrumental in making predictions at greater versus shorter distances? A novel set of diagnostic tools allows us to answer these questions…

Slope aspect is a predictor in many environmental machine-learning models. Here I compare strategies for dealing with problems caused by the cyclic nature of this directional variable.

Partial dependence and ALE plots can be confusing and misleading when interpreting ML models with many correlated features, e.g. in remote sensing. The wiml package implements a new approach that transforms the feature space, but not the training data. How’s that possible? This post walks you through an application of the proposed approach in remote sensing.