Abstract: To improve the accuracy of evaporation duct prediction in the maritime combat environment, machine learning methods are introduced to reconstruct a prediction model of evaporation duct characteristics. It can support radio system design, combat command and control, and seize the advantages of electromagnetic spectrum warfare. The comparative results show that the deviation between the proposed model and the measured profile is 0.45 M-unit, and the relative error is reduced to 0.14%. The root-mean-square error is 1.20 m, which is an increase of 3.46% in accuracy compared to the Gerstoft model. This method is expected to have great potential in predicting the height of the evaporation duct in China and even globally.