Prediction of Wheat Yield in Henan Province Based on xDeepFM Algorithm

Abstract

Grain production has a very important impact on the national economy and people's livelihood. High-precision prediction of grain yield has important practical significance and theoretical research value. Development of latest machine learning technology opens up new way for it. This paper summarized the data related to wheat yield in five provinces: Henan, Hebei, Shandong, Shanxi and Shaanxi, involving 15 influencing factors: such as wheat sown area, the number of primary industry employees, the total power of agricultural machinery, and so on. Then the correlation analysis on the relevant influencing factors was done. And the wheat yield of Henan Province from 2006 to 2020 was predicted based on the xDeepFM deep learning algorithm with good sparse feature processing ability. Wheat predicted yield data for 15 consecutive years have generally performed well. The maximum absolute percentage error is 12.35%, the minimum is 0.35%, and the mean absolute percentage error is 7.71% Some of the poor annual prediction results may be affected by the particularity of local regional factors such as nature or policy in Henan province, resulting in an error of more than 10% in the prediction yield. The results show that the xDeepFM algorithm can be used to predict wheat yield effectively in Henan Province. It has an important reference value for the prediction of provincial grain yield by using xDeepFM model.