Prediction of Soybean Yield in Jilin Based on Diverse Machine Learning Algorithms and Meteorological Disaster Indices

Abstract

To enhance the precision and timeliness of soybean yield prediction in Jilin Province, this study leveraged data from 43 weather stations within the region. Employing methodologies such as Random Forest, Genetic Algorithm-assisted BP Neural Network, Support Vector Machine, and Convolutional Neural Network, predictive models for soybean yield were developed, with a specific focus on comparing the outcomes when including meteorological disaster index variables. The research findings highlight the paramount importance of certain feature variables closely linked to soybean yield, namely minimum temperature, accumulated temperature ≥10°C, mean temperature, frost-free days, maximum temperature, longitude, and precipitation. Notably, the integration of meteorological disaster index variables led to superior simulation results. Among the models utilizing these variables, the Random Forest model demonstrated the highest simulation accuracy, while the GA-BP and SVM models displayed relatively lower performance, with MAE values of 4.45 and 4.49 respectively. Conversely, the CNN model showcased the weakest performance in the context of this study. Ultimately, the collective models exhibited commendable accuracy levels.