Classification of household poverty in West Java using the generalized mixed-effects trees model

Abstract

Dealing with fixed effects and random effects can be accomplished by combining statistical modeling and machine learning techniques. This paper discusses the modeling of fixed effects and random effects using a statistical machine-learning approach. We used the generalized mixed-effects trees (GMET), a tree-based mixed-effect model for dealing with response variables that belong to the exponential family of distributions. In this study, both simulation and actual/empirical data utilized the GMET method to discover data conditions that were appropriate for employing this approach. The simulation data was generated using different response variable generations, as well as different values of the variance of random effect and fixed effect coefficients. The findings indicated that the GMET performs similarly for different response variable generation scenarios. However, it performed better when the fixed effect value and the variance of random effects were large. When applied to the empirical data, the GMET method describes fixed effects and random effects and classifies household poverty status quite well based on the area under curve (AUC) value. It has also revealed that important variables for poverty classification are the number of household members, owning land, the type of main fuel used for cooking, and the main source of water used for drinking. In order to address the socioeconomic disparity that leads to poverty, the government may become concerned about these factors. In addition to that information, the use of regional typology as a random effect in the model has also contributed to the variation of household poverty status. Based on research, the fixed effects in mixed models do not need to be linear and GMET may be employed in grouped data structures, giving the GMET technique the ability to compete with other approaches/methods.