Data sample formation and quality criteria for modelling intelligent warehouse logistics systems

DOI: 10.31673/2412-9070.2026.017401

Abstract

In the modern world, the rapid growth of the e-commerce sector places unprecedented strain on logistics systems, requiring warehouse complexes to be highly adaptable and able to process orders quickly. Traditional approaches to warehouse management, based on fixed storage locations and the classification of goods by sales volume, are ineffective in environments with high demand variability and strict delivery time requirements. The article explores the issue of increasing the productivity of warehouse processes by moving from a purely quantitative increase in physical capacity to the implementation of intelligent data processing algorithms.
The purpose of this work is to determine the prerequisites and tools for algorithmic optimization of the placement of commodity items based on the identification of hidden patterns of joint demand. First, an analysis of current technological trends was conducted, which demonstrated a shift in emphasis from picker-to-parts systems to robotic parts-to-picker systems (MRFS, AGV), the effectiveness of which significantly depends on the quality of routing algorithms and the assignment of storage locations.
With the growth of data volumes, the key factor for success is the formation of a high-quality information base for modelling. The paper examines in detail the criteria for data relevance for logistics tasks, including the presence of accurate time stamps, transactional integrity, and detailed physical parameters of cargo. Particular attention is paid to both the advantages and potential challenges of working with open data sources. A systematic review of eight selected datasets from the Kaggle and UCI Machine Learning Repositories was conducted, and their suitability for solving specific tasks was analysed: from time series clustering and peak load forecasting to packaging optimization.
Separately, the role of internal corporate data of enterprises is highlighted, in particular operational logs of warehouse management systems, which allow taking into account real infrastructure limitations and behavioural factors of personnel. It is concluded that the digital transformation of a modern logistics centre is a synergy of robotic technologies and machine learning methods. The results of the study confirm the feasibility of using clustering for dynamic SKU redistribution, which allows minimizing movement distances and transforming logistics from a model of responding to demand post-facto into a system of planning based on forecasts. 

Keywords: warehouse logistics; optimization of goods placement; data clustering; machine learning; parts-to-picker systems; picker-to-parts systems; e-commerce; big data analysis; demand forecasting.