Web-Based Machine Learning Implementation for Product Demand Prediction at BLE'E Coffee Cakung MSME

Abstract

MSMEs in the culinary sector, such as BLE'E Coffee Cakung, often struggle to manage raw material inventory due to inaccurate product demand estimation, leading to overstocking or stockouts. This research aims to implement machine learning algorithms to build an accurate demand prediction model, compare algorithm performance, and design a web-based prediction information system for business owners. The research uses a Research and Development (R&D) approach with the Waterfall development model, following the CRISP-DM data analysis framework. Linear Regression (Ordinary Least Squares) was implemented natively in PHP as the primary method, with Support Vector Machine (SVM) discussed as a theoretical comparator. Model evaluation used MAE, RMSE, MAPE, and R² metrics; the system was validated through black-box testing and structured user acceptance interviews. Results for the Cold Brew Leci product show excellent accuracy (R² = 0.933; MAPE = 0.28%), a 100% success rate across 20 black-box test scenarios, and a Highly Feasible user acceptance rating (100%). This research provides a practical technology solution for culinary MSMEs in data-driven decision making.