GLMM and GLMM Tree for School Dropout Modeling in Bengkulu Province

Will be published at Friday, 23 May 2025, 10.10 (GMT +7)

Abstract

In machine learning, there are many popular methods for classification and regression modeling, such as the combined tree method. Combined tree methods such as random forests are a combination of many decision trees that tend to have high accuracy, but the resulting models tend to be complicated. Therefore, in modeling that contains random and fixed effects, methods that tend to be simple, such as GLMM, are sometimes more preferred. Currently, many model-based decision tree methods are being developed, such as GLMM trees, which are developed by taking advantage of GLMM and decision trees. This method produces a simpler model than random forests with high accuracy. In this paper, modeling of school dropouts for children aged 6-18 years in Bengkulu Province in 2019 was carried out using GLMM and GLMM Tree with replications. The results of the real difference test on the performance evaluation value showed that GLMM modeling was not significantly different from GLMM Tree. Important variables in the GLMM model are gender, age, KIP, savings ownership, and social assistance. In the GLMM tree model, the important variables are age, disability, savings ownership, and social assistance.