Predictive analytics in smart healthcare for child mortality prediction using a machine learning approach

Area of Science:

• Public Health
• Data Science
• Machine Learning

Background:

• Child mortality remains a critical concern in developing nations, with Pakistan facing a high under-five mortality rate (69/1000 live births).
• Sustainable Development Goal 3 (SDG3) aims to reduce global under-five mortality to 25/1000 live births.
• Predictive analytics offers transformative potential for personalized healthcare and targeted interventions.

Purpose of the Study:

• To develop and evaluate a predictive analytics framework for under-five child mortality in Pakistan.
• To identify key risk factors influencing child mortality using machine learning.
• To assess the performance of various supervised learning classifiers for mortality prediction.

Main Methods:

• Utilized the Pakistan Demographic and Health Survey (2017-2018) dataset.
• Employed multiple imputation for missing data and Information Gain for feature selection.
• Applied Synthetic Minority Over-sampling Technique (SMOTE) for dataset balancing.
• Trained and compared Decision Tree, Random Forest, Naive Bayes, and Extreme Gradient Boosting classifiers.

Main Results:

• Identified critical risk factors: number of under-five children, birth interval, family size, maternal age, age at first birth, antenatal care, breastfeeding, birth size, and delivery location.
• The Random Forest classifier demonstrated superior performance with 93.8% accuracy, 0.964 precision, 0.971 recall, and 0.967 F1-score.
• A functional predictive framework was established to predict child survival status.

Conclusions:

• Predictive analytics, particularly the Random Forest model, is highly effective in predicting under-five child mortality in Pakistan.
• The identified risk factors provide valuable insights for targeted child health interventions.
• The developed framework can significantly support decision-making for child health programs.