Data-driven machine learning algorithm model for pneumonia prediction and determinant factor stratification among...

Area of Science:

• Utilizes machine learning for predictive modeling in child health.
• Focuses on public health and epidemiology in developing countries.

Background:

• Pneumonia is a leading cause of mortality in children under five globally.
• Significant impact on child survival, quality of life, and economic burden.
• Need for data-driven predictive models for early detection and intervention.

Purpose of the Study:

• To develop a machine learning model for predicting pneumonia in Ethiopian children.
• To identify and stratify key determinant factors associated with childhood pneumonia.
• To inform targeted public health interventions for under-five children.

Main Methods:

• Employed machine learning algorithms on the 2016 Ethiopian Demographic and Health Survey dataset (2035 children).
• Data pre-processing included a 4:1 train-test split and tenfold cross-validation.
• Random Forest algorithm selected for optimal performance, evaluated using confusion matrix and feature importance scores.

Main Results:

• Overall pneumonia prevalence was 31.3% among children aged 6-23 months.
• Random Forest model achieved 91.3% accuracy in predicting pneumonia.
• Top predictors identified: health facility visits, child sex, breastfeeding initiation, birth interval, birth weight, parental education, maternal age, and region.

Conclusions:

• Random Forest is an effective model for predicting childhood pneumonia and its determinants.
• Findings support tailored health interventions and proactive childcare strategies.
• Recommends exploring deep learning for enhanced prediction accuracy in future research.