In order to simulate the nonlinear relationship between soil nutrient, fertilizer application rate and crop yield, a fertilization model based on the gradient boosting decision tree (GBDT) algorithm was established taking soil nutrient content and fertilization application rate as input, crop yield as output with the interpolation of the “3414”maize experiment data. Meanwhile the back propagation neural network (BPNN) model, support vector regression (SVR) model and random forest (RF) model were taken as the reference models. The results showed that the average relative error, the mean absolute error and the root mean square error of the GBDT algorithm were 0.46%, 48.7 kg/hm2 and 62.2 kg/hm2 respectively, which are better than the other three algorithms. It indicates that the fertilization model based on GBDT algorithm can achieve higher precision in simulating the relationship between soil nutrient, fertilizer application rate and crop yield, and can be widely applied in the guidance of precision fertilization.