Association Rules in Random Forest for the Most Interpretable Model

Abstract

Random forest is one of the most popular ensemble methods and has many advantages. However, random forest is a "black-box" model, so the model is difficult to interpret. This study discusses the interpretation of random forest with association rules technique using rules extracted from each decision tree in the random forest model. This analysis involves simulation and empirical data, to determine the factors that affect the poverty status of households in Tasikmalaya. The empirical data was sourced from Badan Pusat Statistik (BPS), the National Socio-Economic Survey (SUSENAS) data for West Java Province in 2019. The results obtained are based on simulation data, the association rules technique can extract the set of rules that characterize the target variable. The application of interpretable random forest to empirical data shows that the rules that most distinguish the poverty status of households in Tasikmalaya are house wall materials and the main source of drinking water, house wall materials and cooking fuel, as well as house wall materials and motorcycle ownership.