Interpretable machine learning methods in the analysis of sales drivers on marketplaces

Introduction. Digital marketplaces are complex and rapidly evolving ecosystems where sales outcomes are shaped by the intricate interplay of consumer behavior, sellers’ strategies, and the algorithmic mechanisms of the platforms themselves. To understand which key factors and how their behavior influence sales – amid high-dimensional data, frequent managerial interventions, and significant interdependence among variables – remains a relevant and methodologically challenging task. Purpose. The paper aims at developing and validating a modular, data-driven approach to identifying and analyzing the factors influencing sales volume, taking into account the specific operational characteristics of marketplaces. Materials and Methods. The proposed approach is based on machine learning methods, specifically XGBoost and Random Forest. To ensure adaptability, the models are trained by a rolling window strategy with distribution-sensitive triggers that respond to changes in data distributions. Feature contributions are evaluated with the SHAP (SHapley Additive exPlanations) method. The approach was tested on data collected from the Wildberries marketplace, covering a 36-week period and comprising 56,448 observations across 1,568 product items. Results. The analysis revealed that the most influential factors are related to search visibility and user engagement – particularly the number of queries in the top-100 and the number of product reviews. Price and advertising features also contribute to sales growth, but their impact is less pronounced. The findings further demonstrate that strategies aimed at improving search rankings and increasing user interest have the greatest effect on projected sales volumes. Conclusions. The proposed approach provides a robust analytical foundation for studying digital sales channels and can serve as an effective decision-support tool in the dynamic and highly competitive environment of online marketplaces.