Modeling the Determinants of Attrition in a Two-stage Epilepsy Prevalence Survey in Nairobi Using Machine Learning

Abstract

This study applies machine learning (including random forest, XGBoost, SVM, and Super Learner) to predict attrition in a two-stage epilepsy prevalence survey in Nairobi. Results show high model performance (AUC up to 0.98), identifying key predictors such as proximity to industrial areas, gender, employment, education, household size, and seizure history. The findings guide targeted strategies to improve follow-up rates and inform a predictive tool for future surveys.