Ensemble Machine Learning Approach to Identify Determinants of Suboptimal Measles Vaccine Coverage in Grand Bassa County, Liberia

Abstract

Objectives: Using machine-learning approaches, the researchers aimed to identify and prioritise drivers of incomplete measles vaccination among children aged 12-23 months in Grand Bassa County, Liberia. Design: Cross-sectional research conducted in a community setting. Setting: Five randomly chosen districts in Grand Bassa County, Liberia, containing both urban and rural communities. Participants: 374 caregivers of infants aged 12-23 months, recruited using multistage sampling between October 2024 and February 2025. The response rate was 87.0%. Primary and secondary outcome measures: The primary outcome was MCV2 completion. MCV1 coverage and dropout rates were considered secondary outcomes. The important determinants were found using ensemble machine learning (Random Forest, XGBoost, and LightGBM) with weighted voting. Results: MCV1 coverage was 62.8% (95% CI: 59.8 to 65.8), and MCV2 coverage was 43.6% (95% CI: 40.6 to 46.6), resulting in a 30.6% dropout rate. The ensemble model attained an accuracy of 60.0% (AUC = 0.585, 95% CI: 0.545 to 0.625). The greatest predictors discovered by feature importance analysis were caregiver education (importance=0.156), distance to health facility (importance=0.142), trust in health workers (importance=0.138), and measles knowledge (importance=0.131). Conclusions: Caregiver education, geographic access, provider trust, and knowledge all substantially impact measles vaccination completion. Targeted interventions that address these characteristics might significantly increase vaccination coverage in Liberia and other low-resource countries.