Improved neural network for predicting blood donations based on two emergent factors

Blood donation forecasting is a critical part of blood supply chain management. However, few studies have focused on modeling blood donation with different emergency factors. The purpose of this study was to investigate the effects of different emergency events on blood donation and to build a suitable blood volume prediction model.

The amount of blood donation from 2015 to December 2021 at Beijing Tongzhou District Central Blood Station was selected as the time series data. First, statistical methods were employed to analyze the effect of weather and epidemic factors on blood donation. Second, a hybrid model of SARIMAX and a neural network was built to predict the blood donation in the next two weeks with two factors.

We identified significant differences in blood donations under different emergency conditions and a high correlation between epidemic status and blood donations. In addition, the decision coefficient improved by 60.7%, and the Root Mean Square Error(RMSE) decreased by 1.668 when using the hybrid model of SARIMAX and the neural network, indicating that the model was effective in reducing the prediction error of blood donation.

The hybrid model approach allows long-term forecasting of blood donations under emergency conditions and provides reliable and accurate forecasting results for blood stations up to 2 weeks in advance, facilitating warnings on the blood supply to relevant hospitals and improving hospital treatment rates while reducing blood transportation costs.