Predicting Blood Transfusion Requests using Machine Learning and Time Series in Supply Chain Management

Abstract

Effective blood supply management in healthcare is crucial to meeting patient needs and preventing shortages or wastage. This study presents a comparative evaluation of machine learning and time series forecasting models for predicting daily blood donation and transfusion demand, using real-world data from the Jordanian Blood Bank (2019–2023). Several models, including ARIMA (Autoregressive integrated moving average), SARIMA (Seasonal Autoregressive Integrated Moving Average), Artificial Neural Networks (ANN), Support Vector Regression (SVR), Logistic Regression (LR), and Random Forests (RF) are implemented and assessed. Among time series models, ARIMA showed optimal performance for forecasting both blood donor and transfusion requests with MSE values of 6.63 and 6.66, respectively. ANN outperformed other ML (Machine Learning) models with the lowest MSE (Mean Squared Error) (6.68). Validation using data from Royal Medical Services confirmed model robustness, particularly ARIMA with updated parameters (4,1,5). The findings highlight the importance of integrating data-driven forecasting methods for strategic decision-making in healthcare supply chains.