Enhancing user experience evaluation with a machine learning framework utilizing bayesian modeling adaptive selection and deep learning

Abstract

In today’s competitive e-commerce landscape, understanding and predicting user behavior is essential for improving conversion rates and reducing site abandonment. Traditional methods such as usability testing and behavioral analytics offer limited real-time insight. The integration of Artificial Intelligence (AI),particularly Machine Learning (ML), has enabled more dynamic and data-driven approaches to modeling user intent. This study presents a behavioral prediction  framework that applies ML techniques to detect purchase intent and predict user abandonment based on online shopping patterns. Three classification models were evaluated: Support Vector Machine (SVM), Naïve Bayes (NB), and Recurrent Neural Networks (RNN). Initial experiments with SVM demonstrated strong performance, achieving a training accuracy of 84.09% and a test accuracy of 83.26%, though recall was limited for the minority class. The NB model achieved 77% accuracy but also faced imbalances in recall and precision. Feature selection techniques were implemented to improve model performance, increasing SVM’s training accuracy to 89% and test accuracy to 87%. A real-time abandonment prediction system was developed using an RNN trained on sequential clickstream data, achieving 93% accuracy, 90% precision, 99% recall, and an F1-score of 96%. These results highlight the superior performance of deep learning in modeling sequential user behavior. The findings demonstrate the value of combining feature selection with advanced ML models for purchase intent detection, offering practical strategies to enhance engagement and retention in e-commerce platforms.