The invention belongs to the technical field of carbon emission prediction, and in particular relates to a carbon dioxide emission prediction method comprising the steps of collecting data including the historical CO2 emission, population, GDP per capita, urbanization rate, secondary industry added value proportion, energy consumption structure, energy strength, overall coal consumption, carbon emission strength and total export-import volume; performing non-dimensionalization on the data, computing a gray association degree between each piece of data and the CO2 emission, and screening CO2 emission influence factor indexes input by the model according to the gray association degrees to achieve feature dimension reduction; using a gray prediction model GM(1,1) to predict the screened CO2 emission influence factor index; and using predicted values of the CO2 emission influence factors to serve as model inputs, and then using an improved shuffled frog leaping algorithm to optimize a least square support vector machine model for predicting the CO2 emission. The method provide by the invention has efficient computing performance and excellent global searching ability.