A Method for Predictive Analysis of Platelet Supply

In the realm of medical management, effectively resolving the disparity between blood supply and demand is fundamental to the management of the platelet supply chain. Central to this is the accurate estimation and prediction of blood supply that is instrumental in conserving resources and minimizing expenses. However, time series analyses that rely on single models struggle to capture the intricacies of changing demand structures of blood and are susceptible to external events, which can lead to distorted forecasts. Moreover, the common pitfalls of conventional model evaluation methods, such as their short-term focus and one-sidedness, hinder the accurate assessment of a model's performance in the field of prediction of blood supply.

This paper aims to construct a more comprehensive platelet clinical supply forecasting system by integrating the strengths of various prediction models and enhancing evaluation methodologies.

This paper introduces a novel Decomposition-Combination Prediction Model that leverages X-13ARIMA-SEATS, where the individual components are modeled using ARIMA, TimeGPT, and SNAIVE methodologies. To assess the model's performance, evaluation metrics are meticulously constructed using a rolling window approach coupled with exponential decay weighting, allowing for a more nuanced evaluation through weighted MAPE. The efficacy and robustness of this approach are subsequently validated using platelet data from the Zhejiang Blood Center, providing a rigorous test of the method's practical applicability.

Empirical analysis demonstrates that the decomposition-combination model outperforms the individual ARIMA, Prophet, and TimeGPT models in terms of forecasting accuracy. Furthermore, sensitivity analysis reveals that while the decay factor influences the weighted MAPE results, the overall assessment of model performance remains robust.

The decomposition-combination model adeptly captures the intrinsic characteristics of platelet series, thereby enhancing the model's forecasting accuracy. Concurrently, the rolling window-based weighted MAPE evaluation method effectively discerns the influence of external events and accurately assesses the forecast performance, thereby enhancing the model's generalization capabilities.