Machine Learning Classification of Online Shopper Purchasing Intentions

Abstract

Social and Web Analytics is becoming important towards analyzing online user behavior and data-informed business decisions of others. In the e-commerce domain, understanding customer behavior on shopping websites provides valuable insights into purchasing patterns and user intent. Despite the abundance of user interaction data, accurately predicting whether a visitor will enter the site and complete a transaction remains a critical challenge for online retailers. This study aims to analyze the user’s intentions through machine learning classification to predict purchasing behavior and website abandonment. The UCI Online Shoppers Purchasing Intention Dataset is used, which contains $\mathbf{1 8}$ features like revenue, visitor types, and days of return. Models-Logistic Regression, Random Forest, and XGBoost-are used to test the dataset. Model performance is measured by accuracy, recall, precision, and F1-score, with Synthetic Minority Over-sampling Technique (SMOTE) applied to address severe class imbalance. The results demonstrate that Random Forest can capture complex, non-linear correlations in user behavior, outperforming the other models with the best accuracy and F1score. The findings offer e-commerce platforms insights for improving high-intend customer targeting and reducing cart abandonment rates.