Comparison of Time Series Methods and Machine Learning Algorithms for Forecasting Taiwan Blood Services Foundation's...

Area of Science:

• Operations Research
• Healthcare Management
• Data Science

Background:

• Blood products face supply uncertainty and short shelf lives, leading to significant wastage.
• Hospitals and blood centers struggle with severe blood shortages due to a limited donor pool.
• Accurate blood supply forecasting is crucial to minimize both wastage and shortages.

Purpose of the Study:

• To develop and evaluate efficient forecasting techniques for blood component supply at blood centers.
• To identify the optimal forecasting method for managing blood inventory.

Main Methods:

• Compared time series models (Autoregressive, ARMA, ARIMA, Seasonal ARIMA, Seasonal ESM, Holt-Winters) with machine learning algorithms (ANN, multiple regression).
• Utilized five years of historical blood supply data from the Taiwan Blood Services Foundation.
• Evaluated model performance using error measures to determine the best forecasting technique.

Main Results:

• Time series forecasting methods demonstrated superior performance compared to machine learning algorithms.
• Seasonal Exponential Smoothing Method (ESM) and Autoregressive Integrated Moving Average (ARIMA) models achieved the lowest error measures.
• These models provide reliable predictions for future blood supply.

Conclusions:

• Developed forecasting models can serve as a decision support system for inventory policy at blood banks, donation centers, and hospitals.
• Efficient blood inventory control can be achieved, reducing both blood shortage and wastage.
• The study highlights the effectiveness of specific time series models for blood supply management.