Online Purchasing Analytics and Recommendation System Using K-Medoids Clustering

Abstract: The rise of online shopping has transformed consumer behavior, generating large volumes of data related to preferences, purchases, and browsing activities. Businesses now rely heavily on analytics and recommendation systems to improve customer satisfaction, drive sales, and deliver personalized shopping experiences. Among different recommendation techniques, clustering methods like K-Medoids have gained attention for their reliability in grouping users and products. Unlike K-Means, K-Medoids uses actual data points as cluster centers, making it less sensitive to outliers and more stable in identifying purchasing patterns. The proposed Online Purchasing Analytics and Recommendation System using K-Medoids clustering was designed to analyze user behavior, identify trends, and generate accurate recommendations. Its scope includes collecting and processing purchase history, browsing activities, product details, and demographics. A web-based dashboard was developed to visualize analytics and user clusters, while a recommendation engine provides personalized suggestions. The system, however, is limited to offline model training and excludes deep learning, advanced NLP, or real-time data integration. Experimental research was employed to measure the system’s effectiveness, supported by ISO 25010 quality model-based evaluations. Both qualitative and quantitative tools, including Likert scale questionnaires, ensured systematic, reliable, and meaningful results on attributes like usability, reliability, and functional suitability. The development process followed Agile (SCRUM) methodology, where tasks such as data preprocessing, feature extraction, clustering, and recommendation logic were iteratively built and refined. Regular sprint reviews and team collaboration ensured flexibility and continuous improvement. For pilot testing, Barangay 432 in Sampaloc, Manila, with 2,136 residents, was selected due to its active online shopping culture. Out of this population, 1,856 participants tested the system, alongside businesses, IT students, and faculty members, bringing the total respondents to 1,976. Evaluation results showed high usability (4.13), functional suitability (4.11), and security (4.07), while reliability (3.96) and efficiency (3.94) also performed positively. The overall weighted mean of 4.04 confirmed that users agreed the system met quality expectations, validating its readiness for real-world deployment. In conclusion, the study demonstrates that K-Medoids-based recommendation systems significantly enhance online retail by providing accurate, personalized, and meaningful product suggestions. Retail Insight not only improves user satisfaction and shopping efficiency but also supports better stock management and eco-friendly product promotion. The study recommends expanding the system with hybrid recommendation models, dynamic dashboards, real-time sentiment analysis, and integration with sustainability databases for even greater personalization and environmental impact tracking.