Accurately predicting user purchase behavior remains a fundamental challenge for e-commerce platforms, particularly in categories such as apparel and textiles where exploratory browsing is prevalent. Although prior studies have predominantly relied on purchase intentions or proxy indicators, such measures often inadequately capture realized purchasing outcomes. To address this limitation, this study proposes a deep learning-based framework for session-level purchase behavior prediction that jointly incorporates platform engagement signals and customer attributes. Using large-scale post-filtered, high-intent e-commerce session data characterized by class imbalance, a range of machine learning and deep learning models were systematically evaluated within a unified experimental framework. The empirical results indicate that platform engagement features make a stronger contribution to purchase prediction than customer attributes, while customer attributes provide complementary contextual information. Compared with traditional machine learning approaches, the proposed deep learning model achieved the strongest overall predictive performance under the examined experimental setting. The present study focuses on post-filtered high-intent sessions and should therefore be interpreted as purchase prediction within decision-relevant shopping traffic rather than full-traffic platform prediction. Overall, these findings underscore the effectiveness of jointly modeling short-term behavioral signals and customer-level contextual attributes summarizing historical tendencies in predicting purchase outcomes. From a practical standpoint, the proposed framework facilitates decision support for operational tasks that require accurate session-level purchase prediction in e-commerce platforms. This study thus provides empirical evidence supporting the value of deep learning for modeling complex user behavior and contributes to the advancement of data-driven purchase prediction in online retail environments.

user purchase behavior, deep learning, session-level prediction, customer attributes, textile e-commerce