Rudy Ariyanto, Rakhmat Arianto, Ahmadi Yuli Ananta, Imam Fahrur Rozi, Raynor Herfian Iqbal Fawwaz
The cold-start problem remains a critical limitation in recommender systems, particularly for small businesses where transaction histories are sparse and user preferences are difficult to infer. This study proposes a lightweight recommendation framework for Tuffero bakery that integrates K-Means clustering, user-based collaborative filtering (UBCF), and budget-aware ranking to ensure both stability and affordability of outputs. Transaction data collected between March and December 2023 were enriched with demographic and geographic attributes such as gender, district, location type, and elevation to construct user profiles. Experimental results showed that the system consistently produced the same Top-3 recommendations (macaron mini isi 5, chocopia lumer cokelat, and bites brownies) across three dataset expansions, despite modest silhouette coefficients (0.1140-0.1481). These findings demonstrate that cluster homogeneity combined with budget constraints can compensate for weak structural separation, yielding robust outputs under sparse conditions. The proposed model differs from existing approaches by explicitly incorporating affordability as a filtering mechanism, offering a practical contribution for micro-retail contexts where price sensitivity and limited computational infrastructure are decisive factors. By emphasizing lightweight design, interpretability, and consumer affordability, this study advances clustering-enhanced collaborative filtering toward practical deployment in real-world small enterprise environments. © 2025 IEEE.
Malang State Polytechnic, Department of Information Technology, Malang, Indonesia