RICE YIELD FORECASTING FOR EXPORT PLANNING USING TIME SERIES AND MACHINE LEARNING

Abstract

This independent study aims to apply machine learning techniques to forecast rice demand and supply in Thailand, identify factors influencing rice production, and support decision-making to increase rice sales at CM Premium Rice Co., Ltd. The study employed the CRISP-DM methodology, using data from the Rice Department, the Office of Agricultural Economics, and the company. Five time-series models—ARM, ARIMA, SARIMA, LSTM, and Prophet—were developed using Google Colaboratory. SARIMA achieved the lowest AIC, while Prophet provided the best forecasting accuracy, with the lowest MAE and RMSE of 1,096.78 and 1,422.27, respectively. For decision support, three classification algorithms—Decision Tree (J48), Naïve Bayes, and Support Vector Machine (SMO)—were evaluated. Decision Tree (J48) achieved the highest accuracy of 99.94%. The factors influencing rice production were rainfall, humidity, and temperature. When rainfall and humidity are low, the company should source Thai Hom Mali rice from the upper northern provinces of Thailand. The Prophet model's Lower Bound can support order planning by reducing the risk of stock shortages, while the Upper Bound can help prevent excess inventory. These findings can support the company's sales, inventory, and procurement planning

Keywords: Rice Yield Forecasting, Machine Learning, Time Series Analysis