Explainable Machine Learning for E-Commerce Purchase Prediction Using Session-Level Behavioral Data

This study proposes a behavior-aware explainable machine learning framework for predicting e-commerce purchase intention using session-level behavioral data. The framework integrates predictive modeling, imbalance-aware evaluation, and SHAP-based interpretability within a unified analytical structure. Four supervised learning models, including Logistic Regression, Random Forest, XGBoost, and LightGBM, are evaluated under identical experimental conditions to ensure a fair comparison. Experimental results demonstrate that ensemble-based methods outperform Logistic Regression under imbalanced conditions, particularly in identifying purchasing users. Among ensemble models, LightGBM and XGBoost achieve the highest performance, followed by Random Forest. SHAP-based interpretability analysis reveals that user conversion behavior is primarily driven by behavioral engagement features, with Page Values, ExitRates, and Bounce Rates identified as the most influential predictors, while technical attributes have limited impact. Overall, the findings indicate that combining imbalance-aware learning with structured explainability enables both improved predictive performance and a more transparent understanding of user behavioral decision mechanisms in e-commerce environments.

کلیدواژه ها:

E-commerce conversion prediction ، behavioral analytics ، imbalanced learning ، explainable machine learning ، SHAP-based interpretability ، explainable AI