GLMM and GLMM trees for school dropout modeling in Bengkulu Province

¹ Statistics and Data Science Study Program, School of Data Science, Mathematics, and Informatics, IPB University, Meranti Wing Street, Dramaga Campus IPB, Bogor, West Java, 16680, Indonesia
² Statistics Study Program, Faculty of Mathematics and Natural Sciences, Bengkulu University, Kandang Limun Street, Bengkulu, 38371, Indonesia

^* Corresponding author: khairil@apps.ipb.ac.id

Abstract

The complexity of machine learning models often makes interpretation difficult, despite their high accuracy. This study aimed to model the factors of school dropout among children aged 6-18 in Bengkulu Province in 2019 using a simpler and more interpretable approach. Generalized linear mixed models (GLMM) and GLMM trees were used in this study. GLMM is effective for data with both random and fixed components, whereas GLMM trees is a decision tree-based method that integrates the strengths of GLMM. The performance evaluation results showed no significant difference in accuracy between the GLMM and GLMM trees. The GLMM analysis identified gender, age, ownership of the Indonesia Smart Card (KIP), savings ownership, and social assistance as important variables. The GLMM trees model highlighted age, disability, savings ownership, and social assistance as important variables. In conclusion, both GLMM and GLMM trees are effective in modeling school dropout, but GLMM trees offer a model that can explain the relationships between variables in greater detail.