Demand Forecasting Using Ensemble Learning for Effective Scheduling of Logistic Orders

This study tested gradient-boosted models on daily demand data for nearly 197,000 media products. It compared forecasting by product group and by individual product, including rolling one-week-ahead forecasts. Seasonal and statistical features and information from clusters of similar demand patterns improved predictions. Accuracy varied across product types.

Advances in Artificial Intelligence, Software and Systems Engineering, 271, 313–321