Abstract

Coffee is a vital commodity in Vietnam, providing economic and social significance. Additionally, Vietnamese coffee significantly contributes to global coffee production and is a crucial component of the worldwide coffee supply chain. Price forecasting provides essential information to supply chain members, enabling them to manage their supply chains better, minimize waste, and promote sustainable development. Thus, improving price forecasting accuracy will be beneficial in stabilizing and boosting coffee production by enabling supply chain actors to make informed production decisions, promoting sustainable coffee supply chains, and enhancing Vietnamese coffee competitiveness. However, coffee prices have shifted considerably since the COVID-19 pandemic, making forecasting challenging. As a result, adopting advanced forecasting techniques is critical to enhancing forecasting accuracy. Machine learning (ML), a subset of artificial intelligence (AI), has proven effective and widely used in various fields, particularly forecasting. This study employed a quantitative approach to evaluate the effectiveness of ML models, specifically the Long Short-Term Memory (LSTM) model, in forecasting coffee prices. This study utilized a comprehensive dataset comprising daily global coffee prices for over 50 years. The LSTM model delivered a remarkable forecast accuracy of 98.31%. The experiment results showed that LSTM outperformed Autoregressive Integrated Moving Average (ARIMA) models, one of the most frequently employed classical forecasting methodologies. Moreover, this study demonstrated the effectiveness of ML techniques in forecasting coffee prices, including LSTM, artificial neural networks (ANN), Gradient Boosting, and Random Forest. All ML models achieved excellent forecast accuracy, with values above 98%. LSTM produced the best performance among ML models. Furthermore, this study offers practical solutions to promote the application of ML methods, enhance forecasting accuracy, and improve coffee production efficiency, ultimately developing sustainable supply chains. These solutions involve major stakeholders, including the government, supply chain agents, technology suppliers, and researchers. Collaboration among these stakeholders is critical to successfully installing and optimizing ML forecasting systems.