Design of Demand Forecasting System for Organic Vegetable Supply Using Statistical, Machine Learning, and Deep Learning

Abstract

Organic vegetable supply planning is affected by demand uncertainty, production lead times, and product perishability. This study aimed to design a demand forecasting system for Think Fresh by comparing statistical, machine learning, and deep learning approaches. Weekly demand data for Caisim, Selada Keriting, Pakcoy, Daun Bawang, and Kailan Baby were analyzed using Naive Forecasting, Moving Average, ETS-Damped, CatBoost, and Neural Network models. Forecast accuracy was evaluated using several error metrics. No single model performed best for all commodities. ETS-Damped produced the best result for Caisim with an MAPE of 17.63%, CatBoost for Selada Keriting and Kailan Baby, and NeuralNetwork for Pakcoy and Daun Bawang. The selected models were integrated into the forecasting system to support data-driven production planning at Think Fresh.