Decoding Drop-Off Points: A Multi-Factor Machine Learning Model for Cart Abandonment in Personalization-Heavy E-Commerce Sites

Abstract

In today’s hyper-personalized e-commerce landscape, cart abandonment remains a persistent and complex challenge, with significant implications for revenue and user retention. Despite substantial investments in personalization technologies, e-commerce platforms continue to experience high rates of cart abandonment, signaling a misalignment between personalization strategies and user decision-making processes. This study addresses this gap by developing a multi-factor machine learning model aimed at decoding the underlying triggers that lead users to abandon their shopping carts, particularly within personalization-heavy e-commerce environments. The objective of this research is to identify, quantify, and interpret the most influential behavioral, contextual, and system-level factors contributing to cart abandonment. A diverse dataset from a major e-commerce platform was analyzed, encompassing user demographics, session behavior, product interaction logs, real-time recommendation exposures, and temporal engagement variables. Our modeling approach integrates both interpretable and high-performance machine learning techniques, including Logistic Regression, XGBoost, and LightGBM. Evaluation metrics such as AUC-ROC, F1-score, and precision-recall were used to rigorously assess model performance across multiple validation folds. The results indicate that cart abandonment is significantly influenced by a combination of user-specific and system-level features. The most critical factors included session duration, frequency of personalized recommendations, mobile page load latency, discount volatility, and timing of last cart interaction. Notably, while personalization increased engagement, excessive exposure to irrelevant recommendations often led to decision fatigue and exit behavior. SHAP (SHapley Additive exPlanations) analysis further provided model transparency, highlighting nuanced feature contributions and interaction effects. This study concludes that combating cart abandonment in personalization-heavy contexts requires a shift toward adaptive personalization strategies, real-time friction detection, and more context-aware UX design. The findings offer a strategic framework for e-commerce platforms seeking to optimize conversion rates through data-driven behavioral insights.