FROM CLICKS TO CONVERSIONS: INTERPRETING ONLINE CONSUMER BEHAVIOR USING EXPLAINABLE AI

Due to increased dominance of e-commerce in global retail industry, behavior knowledge makes online consumers engaged to end of buying products or services has become essential to enterprises. Irrespective of growing popularity of machine learning (ML) models predict conversion outcomes, there is a lack of interpretability limiting actionable information. This paper is a thorough report on explanation of online consumer behavior by utilizing two state of art exemplifiable AI (XAI) methodology namely SHAP (SHapley Value Additive exPlanation) and LIME (Local Interpretable Model-agnostic) on centered scores given by a gradient-based escalated tree model constructed on a genuine world data of one million 2021 to 2024 shopping sessions. Study goes through behavioral characteristics, like session length, views per session, referral sources, device types and abandoned carts. According to study results, session length and time on a product page stand out as the most important conversion drivers, and mobile users have a much lower conversion rate in comparison with desktop users, even though level of engagement is higher. SHAP analysis reveals non-linear influence between referral sources and user demographics whereas LIME learns local decision rules explain an individual conversion. In addition, research proposed a new hybrid XAI approach, SHAP-LIME Fusion, uses less explanation instability and outperforms individual methods in 32%. Findings are beneficial to digital marketers and UX designers who can use them to improve conversion funnels. Study finds its significance to both academia and practice of e-commerce since it not only fills in gap between prediction correctness and interpretability in e-commerce analytics.