The data-driven, multi-item, one-warehouse, multiretailer (OWMR) problem is examined by leveraging historical data and using machine learning methods to improve the ordering decisions in a two-echelon supply chain. A deep stacking kernel machine (DSKM) and its adaptive reweighting extension (ARW-DSKM), fusing deep learning and support vector machines, are developed for the data-driven, multi-item OWMR problems with backlog and lost sales. Considering the temporal network structure and the constraints connecting the subproblems for each item and each retailer, a Lagrange relaxation–based, trilevel, optimization algorithm and a greedy heuristic with good theoretical properties are developed to train the proposed DSKM and ARW-DSKM at acceptable computational costs. Empirical studies are conducted on two retail data sets, and the performances of the proposed methods and some benchmark methods are compared. The DSKM and the ARW-DSKM obtained the best results among the proposed and benchmark methods for the applications of ordering decisions with and without censored demands and with and without new items. Moreover, the implications in selecting suitable, that is, prediction-then-optimization and joint-prediction-and-optimization, frameworks, models/algorithms, and features are investigated. History: Accepted by Ram Ramesh, Area Editor for Data Science and Machine Learning. Funding: This work was supported by the National Natural Science Foundation of China [Grant 72371062]. Supplemental Material: The software that supports the findings of this study is available within the paper and its Supplemental Information ( https://pubsonline.informs.org/doi/suppl/10.1287/ijoc.2022.0365 ) as well as from the IJOC GitHub software repository ( https://github.com/INFORMSJoC/2022.0365 ). The complete IJOC Software and Data Repository is available at https://informsjoc.github.io/ .

This study develops machine learning methods for the data-driven demand estimation and assortment planning problem by addressing three subproblems, that is, demand forecasting simultaneously considering cross-selling and substitutions, estimation of the cross-selling and substitution effects, and assortment optimization. These three subproblems are transformed into three sequentially related machine learning problems: collective demand forecasting, demand inference for cross-selling and substitutions, and assortment rule mining. For collective demand forecasting, related product features are introduced to consider both the cross-selling and substitution effects, and a collaborative coordinate descent method with a good convergence property is developed to make distributed demand forecasting and a global update of related product features. Using the results, demand inference adopts transfer and semisupervised learning methods to tackle the challenge of missing data in quantifying the cross-selling and substitution effects. For assortment rule mining, the assortment rules bridge the gap between prediction and optimization, and the developed heuristics obtain the best assortment using the prior knowledge discovered in demand inference. The computational results on a real-world database and a semisynthetic database show that collective demand forecasting obtained far better results than the standard demand forecasting methods and some popular graph learning methods, and the developed heuristics identified much better assortments than those obtained with the baseline methods. History: Accepted by Ram Ramesh, Area Editor for Data Science and Machine Learning. Funding: This work was supported by the construction base project of discipline innovation and talent introduction plan of Chinese higher educational institutions (111 project) [Grant B16009] and the National Natural Science Foundation of China [Grant 72031002]. Supplemental Material: The online appendices are available at https://doi.org/10.1287/ijoc.2022.1251 .

The prevalence of omnichannel marketing and sales enables firms to make ordering decisions based on multisource heterogeneous information from all the channels. This work extends the data-driven inventory management models with single-source or homogeneous information to those with multisource heterogeneous information. Representation learning and information fusion strategies are used to deal with the challenges of high dimensionality and heterogeneity of multisource heterogeneous information. Quantile regression convolutional attention neural networks with different structures embedding different information fusion strategies are developed to address the problems. Two cases with real-world data are studied, and the results show that the proposed methods using representation learning and/or information fusion strategies have (far) better performances than the existing methods without using these strategies. Moreover, some managerial insights into representation learning and information fusion, different from the practices of deep learning in computer vision and natural language processing, are provided for inventory management problems.