Abstract
The digital transformation of global markets has drastically reshaped customer interaction, behavior, and expectations. E-commerce platforms now face increasing pressure to predict consumer actions with accuracy while ensuring transparency in their algorithms. Explainable Artificial Intelligence (XAI) has emerged as a pivotal tool to bridge this interpretability gap. This research aims to integrate XAI techniques into customer behavior modeling to enhance decision-making processes in e-commerce ecosystems. It investigates how interpretable models can identify key behavioral features, predict future actions, and optimize user experience. The goal is to provide strategic insights that can guide personalized marketing, inventory planning, and customer engagement strategies. The study employed supervised machine learning algorithms—specifically decision trees, SHAP (SHapley Additive exPlanations), and LIME (Local Interpretable Model-Agnostic Explanations)—on a dataset of 100,000 anonymized e-commerce user sessions. Key features analyzed included session duration, clickstream depth, bounce rate, and previous purchase history. Data preprocessing, feature engineering, and model tuning were conducted using Python libraries (scikit-learn, SHAP, and LIME). The best-performing model, XGBoost, achieved 89.2% accuracy, 87.6% F1-score, and an AUC of 0.91. SHAP analysis showed that session duration (avg. 420s), search depth (≥6 categories), and purchase history had the highest impact on conversion predictions. Behavioral segmentation revealed that returning buyers had a 78.3% conversion rate, and mobile app users converted at 63.4%, with cart abandoners at only 12.7%. These findings confirm that combining explainability with predictive modeling improves trust, transparency, and usability in business intelligence workflows. Applications include real-time adaptive recommender systems, personalized retention strategies, fraud detection, and churn prediction—supporting more ethical, efficient, and data-driven decision-making across digital commerce platforms.