Variable Importance in Random Uniform Forests

Saïp Ciss (1)

1 MODAL'X - Modélisation aléatoire de Paris X

Random Uniform Forests (Ciss, 2015a) are an ensemble model that use many ran-domized and unpruned binary decision trees to learn data. It is a variant of Random Forests (Breiman, 2001) and in this article, we will focus on how variable importance is assessed in Random Uniform Forests. We provide many measures of variable importance and show how they can help to explain the data and how they can enhance prediction tasks. In Random Uniform Forests, the main purpose of Variable Importance is to assess which, when, where and how covariates have influence on the problem. We provide a description of measures of Variable Importance as they are defined in the model, with full comprehensive examples and many visualization tools. These ones may be viewed as the shadow of Variable Importance techniques and all tools discussed can be found in the randomUniformForest R package.

Mots clés: Random Uniform Forests; Variable importance; R package; regression; ensemble learning; classification; machine learning; statistical learning

Domaines: Mathématiques [math]; Machine Learning [stat.ML]

Saïp Ciss. Variable Importance in Random Uniform Forests. 2015. ⟨hal-01104751⟩