From Basic Indicators To Direct Health Status: A Multi-Model Machine Learning Approach For Malnutrition Detection In Children

Abstract

Malnutrition in children is one of the serious global health concerns, especially after COVID-19. The lower and middle-income countries are the most affected and highly susceptible to malnutrition problems. The inadequate access to nutritious food, healthcare facilities, clean water, intervention programs and early childhood care and education are the main causes of the childhood malnutrition crisis. Malnutrition in children under the age of five affects their physical, cognitive and emotional behaviour and development. Furthermore, it reduces their productivity in the long term, which leads to health-related problems, reduced academic performance and economic hardship. Even though various initiatives and intervention services are introduced and implemented globally, malnutrition continues to affect an overwhelming number of children under the age of five.  The immediate necessities for innovative and data-driven strategies are required to improve the early malnutrition detection, prevention and treatment processes across the world. In this era, machine learning is one of the revolutionary technologies in health care systems to build data-driven decisions and targeted interventions in diagnostic and treatment processes. Machine learning algorithms facilitate to analysis of diverse datasets such as anthropometry features, clinical and developmental factors, socio-economic and environmental indicators to detect the malnutrition health status of the children. The different types of machine learning models were developed, such as linear and non-linear models, ensemble methods and boosting methods using Logical Regression, Decision Tree, Random Forest, Gradient Boosting and XGBoost classification algorithms to classify the early malnutrition health statuses of the children. These ML models were enhanced by the Synthetic Minority Oversampling Technique (SMOTE) to reduce class imbalances and a hyperparameter tuning process to reduce overfitting issues. These techniques were effective in this prediction system to enhance its prediction accuracy. The Gradient Boosting machine learning model outperforms other machine learning models with the highest accuracy.