A hybrid framework of statistical, machine learning, and explainable AI methods for school dropout prediction.

Area of Science:

• Educational research
• Data science
• Public policy

Background:

• Student dropout poses significant challenges in Bangladesh, impacting educational attainment and socio-economic development.
• Addressing student attrition is crucial for improving overall educational system effectiveness.

Purpose of the Study:

• To investigate the multifaceted factors contributing to school dropout among Bangladeshi students aged 6-24.
• To develop and validate predictive models for student dropout using advanced analytical techniques.
• To enhance the interpretability of predictive models for transparent policy recommendations.

Main Methods:

• Utilized data from the 2019 Multiple Indicator Cluster Survey (MICS) for Bangladesh.
• Employed a hybrid approach combining statistical analysis (logistic regression) with machine learning (Random Forest, XGBoost).
• Applied explainable AI techniques (SHAP, LIME) for model interpretability.

Main Results:

• The Extreme Gradient Boosting (XGBoost) model demonstrated superior predictive performance with 94.4% accuracy.
• Key predictors identified include age, sex, completed grade, division, wealth index, and parents' educational background.
• Statistical and machine learning analyses successfully identified significant factors influencing student dropout.

Conclusions:

• The study provides a data-driven foundation for understanding and mitigating student dropout in Bangladesh.
• Findings can inform the development of targeted intervention strategies for policymakers.
• Reducing student dropout rates is essential for enhancing educational outcomes and national development.