Design a model to predict incomplete immunization among Ethiopian children using ensemble machine learning...

Area of Science:

• Public Health
• Computational Biology
• Machine Learning

Background:

• Immunization is a crucial, cost-effective public health strategy globally.
• Ethiopia faces challenges in achieving optimal childhood immunization coverage.
• Understanding factors influencing incomplete immunization is vital for improving child health outcomes.

Purpose of the Study:

• To identify key determinants of incomplete immunization among children aged 0-59 months in Ethiopia.
• To leverage ensemble machine learning techniques for accurate prediction and analysis.
• To inform targeted public health interventions for enhanced immunization coverage.

Main Methods:

• Utilized 16,394 Ethiopian Demographic and Health Survey (EDHS) datasets, split into 80% training and 20% testing sets.
• Employed various ensemble machine learning algorithms including Bagging (Random Forest), Boosting (XGBoost, LightGBM, CatBoost), and Voting ensembles.
• Implemented Stacking models with XGBoost/CatBoost as base learners and other ML algorithms as meta-models, all in Python.

Main Results:

• The bagging meta-estimator + XGBoost voting model achieved the highest performance: 95.94% accuracy, 95.89% F1-score, 94.81% recall, and 97.07% precision.
• The model demonstrated high reliability with a 96% AUC-ROC value and 95.75% cross-validation score.
• Marital status and place of residence were identified as the most significant factors associated with incomplete immunization.

Conclusions:

• Ensemble machine learning effectively identifies critical factors influencing childhood immunization status in Ethiopia.
• Marital status and residence are key demographic factors requiring attention in immunization programs.
• The study's findings offer valuable insights for developing targeted interventions to improve immunization coverage and child health.