Seeing the wood for the trees: predictive margins for random forests

Abstract

Abstract Classification trees and random forests offer a number of attractive features to corpus data analysts. However, the way in which these models are typically reported – a decision tree and/or set of variable importance scores – offers insufficient information if interest centers on the (form of) relationship between (multiple) predictors and the outcome. This paper develops predictive margins as an interpretative approach to ensemble techniques such as random forests. These are model summaries in the form of adjusted predictions, which provide a clearer picture of patterns in the data and allow us to query a model on potential nonlinear associations and interactions among predictor variables. The present paper outlines the general strategy for forming predictive margins and addresses methodological issues from an explicitly (corpus) linguistic perspective. For illustration, we use data on the English genitive alternation and provide an R package and code for their implementation.