A Method for Predictive Analysis of Platelet Supply.

Area of Science:

• Medical Management
• Supply Chain Optimization
• Time Series Analysis

Background:

• Effective platelet supply chain management requires accurate blood supply forecasting to balance demand and minimize costs.
• Traditional single-model time series analyses often fail to capture complex demand shifts and external event impacts, leading to inaccurate blood supply predictions.
• Conventional model evaluation methods have limitations, including short-term focus and one-sidedness, hindering accurate assessment of prediction performance.

Purpose of the Study:

• To develop a comprehensive platelet clinical supply forecasting system by combining diverse prediction models.
• To enhance forecasting accuracy and robustness through improved evaluation methodologies.

Main Methods:

• Introduction of a novel Decomposition-Combination Prediction Model utilizing X-13ARIMA-SEATS.
• Individual time series components modeled using Autoregressive Integrated Moving Average (ARIMA), TimeGPT, and SNAIVE.
• Performance evaluation using a rolling window approach with exponential decay weighting for weighted Mean Absolute Percentage Error (MAPE).
• Validation with platelet data from the Zhejiang Blood Center.

Main Results:

• The decomposition-combination model demonstrated superior forecasting accuracy compared to individual ARIMA, Prophet, and TimeGPT models.
• Sensitivity analysis confirmed the robustness of the model's performance assessment despite variations in the decay factor.
• Empirical validation using real-world platelet data confirmed the model's efficacy.

Conclusions:

• The decomposition-combination model effectively captures inherent platelet time series characteristics, significantly improving forecasting accuracy.
• The rolling window-based weighted MAPE evaluation method accurately assesses forecast performance and discerns external event influences, enhancing generalization capabilities.