Performance of Machine Learning Classifiers in Classifying Stunting among Under-Five Children in Zambia

Area of Science:

• Public Health
• Machine Learning
• Pediatrics

Background:

• Childhood stunting remains a significant global public health challenge.
• Early identification and intervention are crucial for mitigating the long-term effects of stunting.

Purpose of the Study:

• To train and evaluate machine learning (ML) classification algorithms for predicting stunting in children under five in Zambia.
• To assess the performance of various ML models using the 2018 Zambia Demographic Health Survey (ZDHS) dataset.

Main Methods:

• Applied Logistic Regression (LR), Random Forest (RF), Support Vector Classification (SVC), XG Boost (XgB), and Naïve Bayes (NB) algorithms.
• Utilized the 2018 ZDHS dataset, selecting 13 out of 58 features using Random Forest.
• Calibrated predicted probabilities and evaluated models using accuracy, recall, precision, and F1 scores.

Main Results:

• Approximately 34.2% (2327 children) were identified as stunted.
• Random Forest (RF) demonstrated the highest accuracy (79% in testing data).
• Naïve Bayes performed as the least accurate algorithm for stunting prediction.

Conclusions:

• Machine learning models can effectively predict childhood stunting in Zambia.
• Calibrating predicted probabilities enhances ML model performance.
• ML aids in the timely diagnosis of stunting, facilitating prompt intervention development.