An Interpretable Predictive Model of Vaccine Utilization for Tanzania

Area of Science:

• Public Health
• Health Informatics
• Machine Learning

Background:

• Current vaccine stock management relies on outdated census data and simplistic forecasting models.
• Existing models lack insights into factors influencing vaccine demand.
• Optimizing vaccine supply chains is critical for healthcare systems.

Purpose of the Study:

• To develop a state-of-the-art machine learning model for accurate vaccine utilization forecasting.
• To improve upon existing forecasting methods by incorporating novel, relevant data.
• To identify key factors influencing vaccine utilization at the health facility level.

Main Methods:

• Developed a multidimensional machine learning model, specifically a random forest regressor.
• Utilized novel, temporally and regionally relevant vaccine utilization data.
• Predicted bi-weekly vaccine utilization at the individual health facility level.

Main Results:

• The machine learning model achieved a forecasting fraction error of less than two for approximately 45% of facilities.
• The random forest regressor demonstrated an average forecasting fraction error nearly 18 times lower than the existing system.
• The model provided valuable insights into factors affecting vaccine utilization.

Conclusions:

• Advanced machine learning models can significantly enhance vaccine utilization forecasting accuracy.
• This approach offers a powerful tool for optimizing vaccine stock management in developing countries.
• The study highlights the potential of AI and big data in transforming predictive health systems.