Fig. 4From: A machine learning approach to small area estimation: predicting the health, housing and well-being of the population of NetherlandsSTAR model terms for “drinker” indicator.. Because a different model is fitted for each of the 25 GGD regions, we calculate the average term value over these models for a given feature value. Even though interpretation is seen as a strength of statistical models, it is quite non-trivial to interpret such a complex model. Compared to the XGBoost SHAP values below, the STAR model appears to have a similar interpretationBack to article page