PENGGUNAAN ALGORITMA MACHING LEARNING UNTUK PREDIKSI STOK PADA USAHA WARKOP MIE ACEH REZEKI BERSAMA

Abstract

This study designs a stock demand prediction system based on Machine Learning using the n8n automation platform. Sales data for 14 days were collected from Google Sheets, preprocessed, and then used to generate predictions with SARIMA and Random Forest algorithms. The results of both models were compared using MAPE. The system also provides average calculations based on day, weather, and events, and automatically sends stock recommendations via WhatsApp along with graphs generated by hcti.io. The research method employed is Research and Development with the Waterfall model. Black box testing was conducted across 6 scenarios, and all features were found to function validly. The results indicate that SARIMA is more accurate with a MAPE of 38.55%, categorized as fair, compared to Random Forest with a MAPE of 50.25%, categorized as poor. SARIMA was therefore selected as the main model for 7-day ahead forecasting and provides recommendations to increase stock by 30%-40% on peak days. Conclusion: The stock prediction system based on n8n and SARIMA is feasible for supporting inventory decision-making at Warkop Mie Aceh Rezeki Bersama.