Explaining student absenteeism in Morocco using TIMSS 2023 data: a deep learning and explainable AI approach

Abstract

In Morocco, learning equity and academic success are seriously threatened by student absenteeism, especially in important subjects like science, math, and languages. Since absenteeism is one of the primary causes of school dropout, which is a priority in the 2022–2026 Strategic Roadmap for Education Reform, it continues to erode classroom participation despite efforts by educational staff to curb its spread. Using information from the TIMSS 2023 survey, which was given to eighth-grade Moroccan students, this study examines the primary variables linked to perceived student absenteeism. An autoencoder-based dimensionality reduction technique was used to extract latent features that capture crucial patterns across student, teacher, and school-level data to manage the high dimensionality of the TIMSS dataset. To differentiate between schools with high and low perceived absenteeism levels, these features were then used to train several machine learning classification models, such as Support Vector Machine (SVM), Random Forest, and XGBoost. The most significant latent features and their matching original variables were interpreted using SHAP values to improve model transparency. The findings demonstrate how important it is for classroom climate, teacher practices, and institutional factors like class size, teaching strategies, technology support, and teacher expectations to influence absenteeism dynamics. In addition to showing the promise of explainable AI in educational research, this method offers practical insights to guide focused interventions in line with national education priorities.