DATA ANALYSIS AND MACHINE LEARNING PROBLEM FORMULATION IN INVENTORY MANAGEMENT UNDER CENSORED DEMAND
Keywords:
machine learning, inventory management, latent demand, censored demand, time series, data analysisAbstract
The purpose of this study is to analyze the structure of daily supplier price lists and to formulate a machine learning problem for inventory management when direct sales data are unavailable. This setting is typical for wholesale and distribution companies, where the available information reflects assortment composition, prices, and stock balances rather than actual customer demand. Materials and Methods. The empirical basis of the study consists of daily price lists from a large wholesale auto parts supplier covering 113 calendar days of observations. The data were preprocessed by normalizing product records, constructing stable SKU identifiers, and forming daily time series that include stock balances, prices, and product availability indicators. The study applies descriptive statistics, time series analysis, assortment dynamics analysis, and detection of consecutive out-of-stock intervals. Particular attention is paid to distinguishing between two different states: a zero-stock balance for an item still present in the price list and the complete absence of an item from the observed data. Results. The processed dataset contains 133,852 unique stock keeping units. For a representative sample of 5,000 SKUs, 36 consecutive zero-stock intervals were identified. The average duration of an out-of-stock period was 39.3 days, the median duration was 25 days, and the maximum duration reached 113 days. It was found that 91.7 % of such periods lasted at least 3 days, 75.0 % exceeded one week, and 55.6 % continued for 14 days or more. These results show that stockouts are not isolated short-term deviations but a systematic source of demand censoring. Conclusion. The study demonstrates that direct use of inventory balances or sales observations without accounting for censoring may lead to biased demand estimation and poorer replenishment decisions. The proposed problem formulation provides a methodological basis for developing machine learning models aimed at recovering latent demand and supporting inventory management under incomplete data observability.References
1. Silver E.A. Operations Research in Inventory Management: A Review and Critique // Operations Research. 1981. Vol. 29, no. 4. P. 628–645. DOI: 10.1287/opre.29.4.628
2. Axsäter S. Inventory Control. 3rd ed. Cham: Springer, 2015. DOI: 10.1007/978-3-319-15729-0
3. Gutierrez J.C., Polo Triana S.I., León Becerra J.S. Benefits, challenges, and limitations of inventory control using machine learning algorithms: literature review // OPSEARCH. 2025. Vol. 62.
P. 1140–1172. DOI: 10.1007/s12597-024-00839-0
4. Albayrak Ünal Ö., Erkayman B., Usanmaz B. Applications of Artificial Intelligence in Inventory Management: A Systematic Review of the Literature // Archives of Computational Methods in Engineering. 2023. Vol. 30. P. 2605–2625. DOI: 10.1007/s11831-022-09879-5
5. Микрюков А.А., Гилёв Д.В. Комбинированный метод прогнозирования спроса на продовольственные товары на основе ансамбля LSTM-сети и SARIMA-модели для интеграции в ERP-системы российского ритейла // Вестник ЮУрГУ. Серия «Компьютерные технологии, управление, радиоэлектроника». 2025. Т. 25, № 4. С. 122–128. DOI: 10.14529/ctcr250409
6. Bertsimas D., Kallus N. From Predictive to Prescriptive Analytics // Management Science. 2020. Vol. 66, no. 3. P. 1025–1044. DOI: 10.1287/mnsc.2018.3253
7. Ban G.-Y., Rudin C. The Big Data Newsvendor: Practical Insights from Machine Learning // Ope¬rations Research. 2019. Vol. 67, no. 1. P. 90–108. DOI: 10.1287/opre.2018.1757
8. Микрюков А.А. Концепция и реализация нового подхода к прогнозированию продаж товарных запасов с использованием методов машинного обучения // Современная наука: актуальные проблемы теории и практики. Серия: Естественные и технические науки. 2025. № 5-1.
С. 115–119.
9. Cao Y., Shen Z.-J.M. Quantile forecasting and data-driven inventory management under nonstationary demand // Operations Research Letters. 2019. Vol. 47, no. 6. P. 465–472. DOI: 10.1016/j.orl.2019.08.008
10. Chen B. Data-Driven Inventory Control with Shifting Demand // Production and Operations Management. 2021. Vol. 30, no. 5. P. 1365–1385. DOI: 10.1111/poms.13326
11. Supervised learning for integrated forecasting and inventory control / J.F. van der Haar,
A.P. Wellens, R.N. Boute, R.J.I. Basten // European Journal of Operational Research. 2024. Vol. 319, no. 2. P. 573–586. DOI: 10.1016/j.ejor.2024.07.004
12. Deep reinforcement learning for inventory control: A roadmap / R.N. Boute, J. Gijsbrechts,
W. van Jaarsveld, N. Vanvuchelen // European Journal of Operational Research. 2022. Vol. 298, no. 2. P. 401–412. DOI: 10.1016/j.ejor.2021.07.016
13. Inventory – forecasting: Mind the gap / T.E. Goltsos, A.A. Syntetos, C.H. Glock,
G. Ioannou // European Journal of Operational Research. 2022. Vol. 299, no. 2. P. 397–419. DOI: 10.1016/j.ejor.2021.07.040
14. de Castro Moraes T., Yuan X.-M. Data-Driven Solutions for the Newsvendor Problem: A Systematic Literature Review // Advances in Production Management Systems. Artificial Intelligence for Sustainable and Resilient Production Systems. IFIP Advances in Information and Communication Technology. Springer, 2021. P. 149–158. DOI: 10.1007/978-3-030-85910-7_16
15. Deep Controlled Learning for Inventory Control / T. Temizöz, C. Imdahl, R. Dijkman
et al. // European Journal of Operational Research. 2025. Vol. 324, no. 1. P. 104–117. DOI: 10.1016/j.ejor.2025.01.026
