Scholarlink

Coffee plays a pivotal role in global trade, with its financial market far exceeding its physical transactions. Accurate forecasting of coffee prices is crucial for stakeholders to make informed decisions. This paper explores the application of machine learning, specifically the Facebook Prophet model, for predicting coffee prices. Leveraging historical data from the U.S. (primarily Arabica coffee) and London (Robusta coffee from Asia and Africa) markets, we resample data to monthly averages for enhanced modelling efficiency. Our study includes comprehensive data processing, visualisation, and time series decomposition to unravel the cyclic nature of coffee prices. We perform stationarity checks using rolling statistics and the Dickey-Fuller test, followed by differencing to achieve stationarity. The Prophet model’s performance is evaluated based on its accuracy in forecasting short-term price trends, achieving a Mean Absolute Percentage Error (MAPE) of an approximate of 10%. Our findings underscore the effectiveness of Prophet model in delivering precise forecasts, supporting strategic decision-making in the coffee industry.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Forecasting Coffee Prices with Facebook Prophet: A Machine Learning Approach

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Tri-Cong Pham,
  • Thi-Luot Nguyen,
  • Van-Dung Hoang,
  • Van Duy Nguyen

摘要

Coffee plays a pivotal role in global trade, with its financial market far exceeding its physical transactions. Accurate forecasting of coffee prices is crucial for stakeholders to make informed decisions. This paper explores the application of machine learning, specifically the Facebook Prophet model, for predicting coffee prices. Leveraging historical data from the U.S. (primarily Arabica coffee) and London (Robusta coffee from Asia and Africa) markets, we resample data to monthly averages for enhanced modelling efficiency. Our study includes comprehensive data processing, visualisation, and time series decomposition to unravel the cyclic nature of coffee prices. We perform stationarity checks using rolling statistics and the Dickey-Fuller test, followed by differencing to achieve stationarity. The Prophet model’s performance is evaluated based on its accuracy in forecasting short-term price trends, achieving a Mean Absolute Percentage Error (MAPE) of an approximate of 10%. Our findings underscore the effectiveness of Prophet model in delivering precise forecasts, supporting strategic decision-making in the coffee industry.