Modeling Zero-Dose Children in Ethiopia: A Machine Learning Perspective on Model Performance and Predictor Variables

Area of Science:

• Public Health
• Machine Learning
• Epidemiology

Background:

• Childhood vaccination remains a challenge in low- and middle-income countries, with many children unvaccinated.
• The Immunization Agenda 2030 aims to reduce unvaccinated children by identifying at-risk populations.
• Effective strategies for identifying and reaching unvaccinated children are limited.

Purpose of the Study:

• To develop a machine learning (ML) model to predict zero dose (ZD) children in Ethiopia.
• To identify the most influential predictors associated with ZD status in Ethiopian children.

Main Methods:

• Supervised ML algorithms were applied to National Immunization Evaluation Survey data of 13,666 children aged 12-35 months.
• Model performance was evaluated using accuracy, AUC, precision, recall, and F1-score.
• Shapley Additive analysis was used to determine the importance of predictor variables.

Main Results:

• Light Gradient Boosting Machine (LGBM) demonstrated superior performance with 93% accuracy, 97% AUC, 94% precision, and 91% recall.
• Significant predictors of ZD status included poor perception of vaccination benefits, lack of antenatal care, and distance to immunization services.
• Other key predictors were absence of maternal tetanus toxoid vaccination and limited access to healthcare.

Conclusions:

• ML models, particularly LGBM, can effectively predict children at risk of being ZD, guiding targeted public health interventions.
• Addressing predictors such as immunization access, maternal care utilization, and awareness of vaccination benefits is crucial.
• Interventions focusing on enhancing maternal care, caregiver education, and improving immunization accessibility can reduce ZD prevalence and vaccination inequities.