RuleFit implements the learning method and interpretational tools described in Predictive Learning via Rule Ensembles (pdf). RuleFit3 is a newer version with improved analytics and some additional options. A principal improvement is the use of glmnet (Friedman, Hastie, and Tibshirani 2008) to perform rule fitting . This provides the full sprectum of elastic net procedures from the lasso to ridge regression, thereby allowing the user to decrease the sparsity (increase the number of terms) in the final model from that provided by the lasso. Sparser models than the lasso can be obtained by choosing forward stepwise or forward statewise rule fitting. Also a different method for cross-validated model selection is implemented that usually results in more accurate models and more honest (less optimistic) estimates of future prediction error. This is especially the case for smaller training data sets. There is also a wider variety of model selection criteria from which to choose for selecting the number of terms in the final model.

References in zbMATH (referenced in 1 article )

Showing result 1 of 1.
Sorted by year (citations)

  1. Marjolein Fokkema: pre: An R Package for Fitting Prediction Rule Ensembles (2017) arXiv