Data-driven analysis of climate impact on tomato and apple prices using machine learning

Machine learning has been used in various areas, but there are few studies on price prediction for agricultural products. Here, a machine learning technique for the price prediction of tomato and apple fruits was attempted based on environment and price data for 12 years. The goal of this study is to discover 1) how much can we accurately predict the product prices with the environmental factors and 2) how much each environmental factor affects to the product prices. This study assumes that the environmental factors directly affect crop growth and thus indirectly determine fruit production and accompanying price. In addition, it is assumed that there are two kinds of time lags, between the change in the environmental factors and their effects on the crop growth, and between the change in the crop growth and its effect on the price. In the process, machine learning techniques were used instead of econometric models commonly used in agricultural economics. The relationship between the environmental factors and fruit price with varying time lags in data-driven manner using long short-term memory (LSTM) was modeled in this study. The study empirically revealed that there are suitable time lags between the environmental factors and fruit price in the price prediction, and taking these time lags into the prediction improves the accuracy. Moreover, the importance of each of the environmental factors on the price using shapely additive explanations (SHAP) was demonstrated though this study, which assists the decision-making process in agriculture against the climate change.