ABSTRACT Accurate demand forecasting for high‐value medical consumables is important for improving inventory planning and reducing supply uncertainty in healthcare systems. However, forecasting remains challenging because demand is influenced by time sensitivity, volatility, and bundle‐driven usage patterns among consumables. This study proposes a Neural Prophet–Association Rule Mining (ARM) forecasting framework that integrates product bundle mining with deep time‐series prediction to improve short‐ and long‐term demand forecasting. First, ARM is used to identify strongly associated consumable bundles from historical procurement transactions and transform these relationships into bundle‐effect forecasting features. Second, these features are incorporated into a Neural Prophet model together with autoregressive lags, trend, seasonality, events, and covariates to capture both temporal demand patterns and inter‐product dependencies. Experiments were conducted using procurement data comprising 3215 projects and 15,874 transactions across multiple categories of high‐value medical consumables. Results show that the proposed model consistently outperformed benchmark methods, including ARIMA, LSTM, Linear Regression, and FB Prophet. For diluent syringes, the model achieved a 7‐day forecasting wMAPE of 0.123, RMSE of 141.281, and WAPE of 0.102, while maintaining superior performance across longer forecasting windows, including 30 and 90 days. Statistical validation and repeated‐run analysis further confirmed the robustness of these gains. These findings show that integrating bundle‐effect features with deep time‐series forecasting improves predictive accuracy and provides useful decision support for medical consumables planning and inventory management.
Kai et al. (Sun,) studied this question.