MACHINE LEARNING APPROACH TO AGRICULTURAL PRICE FORECASTING

Abstract

The use of machine learning is on a rise in many industries, including the agricultural sector. For example, McCormick and Company is a leading spice manufacturer, and they would like to forecast changes in the price of raw spices. The most commonly used spice in North America is black pepper. In this research, done collaboratively with McCormick and Company, we propose a method to predict the price of black pepper ahead by one and three months. Our data-set is made up of Mintec crop prices, OECD financial data, and weather data from the Weather Underground. We explored classic time series models such as the ARIMA and persistence models. They performed poorly, so we turned to linear and support vector regression. We were able to achieve a mean average precision (MAPE) of 3.67 for the linear SVR one month ahead prediction. Empirical results show which features are most informative for predicting black pepper prices.