Product Category Level Demand Forecasting Based on LRFM–CLV Customer Segmentation in B2B E-Marketplace Systems

Accurate demand forecasting poses a major challenge in B2B e-marketplace systems due to the heterogeneous, complex, and nonlinear nature of transaction data. This study introduces a customer-segmentation-based demand-forecasting method that employs tailored machine learning models for each cluster. Customers were first grouped by transaction behavior characteristics, and then several prediction models, namely Random Forest, Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and an averaging ensemble of these models, were evaluated within each cluster using metrics such as Mean Absolute Error (MAE), Symmetric Mean Absolute Percentage Error (SMAPE), Root Mean Square Error (RMSE), and the coefficient of determination (R²). The results showed that no single model outperformed the others across all clusters. LightGBM yielded the best results in most clusters, with R² values of 0.8202 in Cluster 0, 0.7505 in Cluster 1, and 0.6816 in Cluster 2, whereas the averaging ensemble model excelled in Cluster 3 with an R² of 0.9849. Visualization of positively growing product categories highlighted notable differences in demand patterns between clusters. These insights suggest that segmentation-based and contextual modeling approaches can enhance both the accuracy and interpretability of demand predictions, improving support for operational decisions and supply chain planning in B2B e-marketplaces.