Based on the correlation between ground neutron detection and the cosmic ray environment, a dataset was constructed using the detection data of geostationary operational environmental satellites and various neutron detection stations worldwide for the solar activity quiet period. Models for inverting the cosmic ray proton environment from ground neutron detection data were established based on the extreme gradient boost decision tree (XGBoost) and artificial neural network. They use genetic algorithm to solve the optimal hyperparameter and train the parameters of each neuron of the neural network to realize the inversion of the cosmic ray proton environment. The mean square error of the model training is 0.499, and the average inversion error of the test set is 26.9% respectively. Compared with the radiation environment model commonly used in aerospace, the error is usually within 200%, which is significantly improved. Multiple other machine learning algorithms, including support vector regression, error back propagation training, long short-term memory network, were compared and the results show that the model established in this paper has the advantages of short training time, fast computation speed, and low resource consumption.