<![CDATA[The rapid expansion of e-commerce has increased the difficulty of guiding users toward relevant products, particularly as catalogs grow and user preferences become more diverse. This paper presents an end-to-end implementation of a personalized product recommendation feature using a memory-based collaborative filtering approach integrated into an e-commerce platform. Development followed a Rapid Application Development (RAD) workflow, enabling iterative prototyping, integration, and testing of the recommendation module within the operational system. Recommendations were generated using a K-Nearest Neighbors method with cosine-based similarity to identify related items from user interaction histories and to produce Top-N product suggestions in the storefront interface. Model evaluation employed a transactional dataset commonly used for recommender experiments, which was refined from 541,909 records (8 attributes) to 406,829 interaction-focused records (CustomerID, Description, Quantity). Performance was assessed using MAE, RMSE, and F1-score, yielding values of 0.6, 0.8, and 0.6, respectively. The results indicate that collaborative filtering can provide moderately accurate and relevant recommendations when interaction history is available, while also exposing practical limitations for users with limited transactions, reflecting a cold-start constraint. These findings suggest that RAD-supported integration of collaborative filtering is feasible for e-commerce personalization and provides a baseline for further enhancement.]]>
Copyrights © 2026
