Soybean Price Prediction Based on Random Forest Model

Abstract

Accurate soybean price prediction is critical for stabilizing the global food supply chain. However, traditional forecasting methods struggle to capture the multidimensional, nonlinear characteristics of soybean prices and often overlook time lag effects. This study focuses on U.S. soybean export prices and their influencing factors from December 2017 to December 2025, constructing a prediction model using the random forest algorithm. Instead of traditional lag operators, this study machine learning to build 1-, 2-, and 3-month lag features. Combining current and lagged values of five key explanatory variables, research form a 20-dimensional quantitative feature matrix. This study selects the widely used ARIMA model as a benchmark and evaluate both models using four metrics: MAE, MSE, RMSE, and R 2 . Results show the random forest model achieves an overall prediction accuracy of 94.2% (R2=0.942), with MAE=16.82, MSE=465.26, and RMSE=21.57—outperforming the ARIMA model (MAE=38.65, MSE=2045.75). This model enables robust fitting and prediction of multidimensional, multi-lag soybean prices, providing a reliable reference for major soybean-producing countries. Future work can expand data scope and integrate more algorithms to further improve accuracy.