Authors
Suleiman Ali Alsaif
Published in
Frontiers in artificial intelligence. Volume 9. Pages 1936360. Epub Oct 09, 2026.
Abstract
Demand forecasting has been shown to be one of the key facilitators of successful supply chain and inventory management. In this work, we introduce a benchmarking pipeline for demand forecasting in retail that includes data preprocessing, comparative benchmarking of different machine learning/deep learning (ML/DL) approaches, and decision-focused evaluation. Unlike other works that compare predictive performance of ML/DL models and end there, our pipeline links reproducible ML/DL benchmarking and decision-focused evaluation through cost of inventory management in order to derive costs from forecasting error/bias. We also present a scalability study on another retail dataset. Several standard forecasting models, such as Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), CNN-LSTM, and XGBoost, are compared using a unified experimental pipeline on benchmark retail datasets. The experimental results reveal that while the deep learning models efficiently identify temporal dependencies, tree-based ensemble algorithms, especially XGBoost, consistently deliver better results in terms of forecasting accuracy on structured retail demand datasets. A scalability case study on the Rossmann Store Sales benchmark evaluates the top-performing model on a larger, richer dataset. Moreover, the study shows the connection between forecast accuracy and the subsequent inventory performance by performing a cost-based evaluation of inventory outcomes of forecasts made using various models.
PMID:
42920761
Bibliographic data and abstract were imported from PubMed on 24 Oct 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 21
- Comments 0