Probabilistic Feature Selection for Interpretable Random Forest Model

Abstract

It is important to understand the reason behind the machine learning model predictions. In this paper we introduce a new model using random forest technique called Interpretable Random Forests. In this, we make some changes in feature selection technique for some business problems. The probabilistic method chosen to select each feature at every node along with the normal feature selection criteria will give us an Interpretable random forest model which will be compared to the general random forest model that gives an idea of how feature selection impacts the output from a model. Our experimental results shows IRF model has more or same accuracy than RF model and also interpretable.