Explainable E-Commerce Transaction Prediction Using LightGBM and Permutation Importance

This study develops an explainable machine learning framework for predicting e-commerce transaction volumes by integrating Light Gradient Boosting Machine (LightGBM), Permutation Feature Importance (PFI), and Partial Dependence Plots (PDP).Using 120000 daily product-level transaction records from Taobao between January 2024 and March 2025, we develop an explainable machine learning framework for transaction volume prediction., the proposed model achieves R2=0.873, RMSE = 23.89, and MAPE = 0.120, outperforming benchmark models—Random Forest, XGBoost, CatBoost, and Linear Regression—with statistically significant improvements (p < 0.05). PFI highlights platform-guided traffic (Taobao and Zhituoche views) and add-to-cart behaviors as the strongest predictors, while PDP reveals nonlinear thresholds in exposure intensity that amplify transaction probability, particularly for high-brand products. The framework addresses a critical gap in current literature—the absence of globally interpretable yet high-performing transaction-prediction models—and provides actionable insights grounded in Signaling Theory and the Resource-Based View. By combining predictive accuracy, interpretability, and operational scalability, this approach supports transparent, data-driven decision-making in real-time e-commerce environments.