A hybrid neural network based model for blood donation forecasting.

Area of Science:

• * Computational intelligence
• * Health informatics
• * Predictive analytics

Background:

• * Blood stations face challenges in managing whole blood inventory due to fluctuating hospital demand and limited resources.
• * Accurate prediction of daily blood donations is crucial for effective resource allocation and ensuring adequate blood supply.
• * Existing forecasting methods may not adequately capture the complex factors influencing blood donation patterns.

Purpose of the Study:

• * To develop and evaluate a hybrid neural network-based model for predicting daily whole blood donations.
• * To enhance the accuracy of blood donation forecasting at the Beijing Tongzhou District Central Blood Station.
• * To provide a tool for managers to optimize blood resource allocation under variable demand.

Main Methods:

• * A hybrid model, SARIMAX-TCN-LSTM, combining time-series analysis and deep learning (Temporal Convolutional Network and Long Short-Term Memory) was developed.
• * Daily whole blood donation data from January 1, 2015, to November 14, 2021, were utilized.
• * Meteorological and epidemic factors were incorporated as additional predictors to improve forecasting accuracy.

Main Results:

• * The hybrid SARIMAX-TCN-LSTM model significantly outperformed traditional time-series forecasting methods.
• * The model demonstrated twice the fitting effectiveness compared to the baseline and a 33% reduction in Root Mean Squared Error (RMSE).
• * An ablation experiment validated the model's structural effectiveness and improved prediction accuracy.

Conclusions:

• * The developed hybrid neural network model effectively improves the prediction of daily blood donations.
• * This intelligent forecasting approach assists blood bank managers in addressing sudden demand surges.
• * The model contributes to optimizing blood resource management and allocation strategies.