Optimizing Purchase Intention Prediction in E-Commerce

This study investigates the effectiveness of deep learning models—Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Bidirectional LSTM—in predicting purchase intention on e-commerce platforms using session-level data. The dataset, sourced from the UCI Machine Learning Repository, comprises 12,330 sessions with features such as ‘PageValues,’ ‘BounceRates,’ ‘ExitRates,’ and user behavior attributes like ‘Month’ and ‘VisitorType.’ After preprocessing categorical variables using Label Encoding and standardizing numerical features, the data was split into an 80-20% training and testing set. Each of the five models was then trained and assessed using metrics such as accuracy, ROC-AUC, and other classification measures. Among the models, the CNN achieved the highest performance with an accuracy of 90.34% and a ROC-AUC score of 0.9156, followed by the LSTM model with an accuracy of 89.72% and a ROC-AUC score of 0.9102. Although the RNN model performed well, it showed relatively lower metrics with an accuracy of 88.40% and a ROC-AUC of 0.8934. ‘PageValues’ exhibited the strongest positive correlation with purchase intention (Pearson correlation: 0.4926, Spearman correlation: 0.6264), while ‘ExitRates’ and ‘BounceRates’ had significant negative correlations of -0.2071 and -0.1507, respectively. This study identifies CNN and LSTM as robust models for predicting purchase intentions in e-commerce, with ‘PageValues’ being the most influential predictor. The results provide valuable insights for optimizing e-commerce platforms to improve conversion rates. Future research will focus on external validation and expanding the dataset to ensure broader applicability across various e-commerce contexts.