Application of the decision tree method for predicting the yield of spring wheat

Abstract

The results of the development of predictive models of the yield of spring wheat based on the use of the decision tree method are presented. When constructing the models, qualitative factors were taken into account (the level of intensification, the system of soil cultivation, the placement of the crop after steam) and agrometeorological resources (the sum of active air temperatures, precipitation). The minimum number of input parameters (public data) was used for the generality of the system and its versatility for different natural and agricultural conditions. The efficiency of using decision trees for forecasting wheat yield is shown. The accuracy of the constructed models was evaluated on the training and test samples and the following indicators were achieved (CART method): average absolute error - 3.455 (training sample) and 4.446 (test sample); determination coefficient - 0.895 (training sample) and 0.811 (test sample). A set of rules has been obtained that determine the level of crop yield depending on the complex of control actions and the prevailing conditions.