PREDICTING WEBSITE USER ENGAGEMENT USING MACHINE LEARNING: A BEHAVIORAL AND DESIGN FEATURE ANALYSIS

Abstract

User engagement behavior are essential in predicting sales and profit for an organization. This study presents a machine learning-based framework for predicting website user engagement levels using behavioral and design-related features extracted from the Online Shoppers Purchasing Intention Dataset obtained from the UCI Machine Learning Repository. The dataset comprises 12,330 instances and 18 features representing user session characteristics, page interaction metrics, and temporal attributes. Four supervised machine learning models were developed and evaluated: Logistic Regression, Random Forest, XGBoost, and a one-dimensional Convolutional Neural Network (CNN). Experimental results derived from actual model runs demonstrate that XGBoost achieved the highest ROC-AUC of 0.929, followed by Random Forest (AUC = 0.918), CNN (AUC = 0.906), and Logistic Regression (AUC = 0.865). Feature importance analysis identified PageValues, ProductRelated_Duration, and ExitRates as the most influential predictors. The findings provide actionable insights for optimizing website design and content strategy to improve user retention and conversion rates