G.K. Strugach1
1 Financial University under Russian Government (Moscow, Russia)

1 gkstrugach@fa.ru

This paper proposes a hybrid decision support system for managing supplies of automotive spare parts on e‑commerce marketplaces under logistics capacity constraints. The study is motivated by the complexity of distributed fulfillment networks, where individual SKUs exhibit intermittent and seasonal demand, heterogeneous dimensions, and varying profitability across delivery clusters, which limits the effectiveness of traditional average‑based planning approaches.

The core contribution is the integration of machine learning demand forecasting with linear programming-based inventory allocation within a unified «predict‑then‑optimize» framework. On the forecasting side, six algorithms are compared using historical data from 5,725 orders on the Ozon marketplace for June–September 2025, with Random Forest demonstrating the best performance (MAE = = 2,41 units per week, R 2= 0,687). On the optimization side, a linear programming model allocates shipment volumes across SKU-cluster pairs subject to a total weight constraint, unit margins, and logistics costs, thereby maximizing expected profit.

The experimental design comprises four scenarios: a naive baseline without optimization, optimization only, machine learning only, and the full hybrid system. Under a capacity limit of 200 kg, the optimization component increases profit by 15,7%, the machine learning component by 0,8%, while the full hybrid system yields a 20,6% gain relative to the baseline, implying a 4,1% positive synergy effect. Additional experiments reveal a monotonic decline in the relative impact of optimization as the capacity constraint is relaxed, with the profit uplift decreasing from 54% at 100 kg to 1% at 350 kg. From a practical standpoint, the proposed approach enables marketplace sellers to move from intuitive or average‑based replenishment to systematic, data‑driven supply management that explicitly accounts for capacity limitations. The model also provides a quantitative decomposition of the contributions of forecasting and optimization, thereby supporting more informed decisions regarding model deployment and logistics strategy adjustments.

Strugach G.K. Hybrid approach to supply management on marketplaces: Integration of machine learning algorithms and mathematical programming // Neurocomputers. 2026. V. 28. № 4. P. 88–96. DOI: https://doi.org/10.18127/j19998554-202604-09

1. Vishnevskij V.M. i dr. Metody mashinnogo obucheniya v reshenii zadachi prognozirovaniya sprosa na otdel'nye vidy tovarov. T-Comm – Telekommunikatsii i transport. 2024. T. 18. № 10. S. 34–43. (in Russian)
2. Gertsenberger E.A. Prognozirovanie sprosa v sfere elektronnoj kommertsii na osnove ob''edineniya kolichestvennykh i kachestvennykh dannykh. Biznes. Obrazovanie. Pravo. 2025. № 2 (71). S. 141–146. (in Russian)
3. Boldycheva A.G. Analitika bol'shikh dannykh dlya dinamicheskogo planirovaniya i upravleniya zapasami. Auditor. 2024. № 11. S. 38–49. (in Russian)
4. Kostioglo V.Kh., Titovets M.I., Titovets E.I. Proektirovanie i upravlenie tsepyami postavok v elektronnoj roznichnoj torgovle. Ekonomika: vchera, segodnya, zavtra. 2024. T. 14. № 4A. S. 806–815. (in Russian)
5. Nikitin M.A. Rol' logisticheskikh operatsij v optimizatsii finansovykh izderzhek pri osushchestvlenii torgovoj deyatel'nosti na marketplejsakh. Vestnik Altajskoj akademii ekonomiki i prava. 2024. № 8. S. 305–309. (in Russian)
6. Rogulin R.S. Obzor metodov upravleniya tsepochkami postavok. Nauchnyj zhurnal NIU ITMO. Seriya «Ekonomika i ekologicheskij menedzhment». 2023. № 2. S. 164–179. (in Russian)
7. Kochegurova E.A., Sajbert S.M., Tat'yankina K.V. Optimizatsiya parametrov gibridnogo algoritma prognozirovaniya s ispol'zovaniem ansamblya modelej v rezhime real'nogo vremeni. Izvestiya Tomskogo politekhnicheskogo universiteta. Promyshlennaya kibernetika. 2024. T. 2. № 4. S. 26–33. (in Russian)
8. Sergeev V.I., Sergeev I.V. Trendy tsifrovoj transformatsii tsepej postavok. Logistika i upravlenie tsepyami postavok. 2021. № 6. S. 3–8. (in Russian)
9. Alyunov A.N., Mosolova K.D. Prognozirovanie ob''emov potrebleniya elektricheskoj energii s primeneniem metodov mashinnogo obucheniya. Nejrokomp'yutery: razrabotka, primenenie. 2023. T. 25. № 5. S. 58–70. DOI: 10.18127/j19998554-202305-08. (in Russian)
10. Sinitsyn I.N., Titov Yu.P. Optimizatsiya giperparametrov v CALS sistemakh metodom murav'inykh kolonij na primere prognoza ob''emov aviatsionnykh passazhiro- i gruzoperevozok. Sistemy vysokoj dostupnosti. 2025. T. 21. № 1. S. 12–24. DOI: 10.18127/j20729472-202501-02. (in Russian)
11. Gulyaev A.V. Gibridnaya model' glubokogo obucheniya dlya prognozirovaniya znachenij vremennykh ryadov. Nejrokomp'yutery: razrabotka, primenenie. 2025. T. 27. № 3. S. 62–72. DOI: 10.18127/j19998554-202503-08. (in Russian)
12. Elmachtoub A.N., Grigas P. Smart «predict, then optimize». Management Science. 2022. V. 68. № 1. P. 9–26.