Based on the accurate prediction of carbon emissions from road traffic, intelligent transportation systems and green mobility will be used for data-driven decision-making. Given the shortcomings of the old way of dealing with complex multi-variate non-linear relationships and missing data, we propose a new GAN-KAN hybrid model in this paper. Fill in the missing values of the time-series traffic data in the model, and use a function-structure decomposition method to model multivariate carbon emissions more accurately and understandably. Based on the experimental results, the enhanced GAN increased the number of samples from 18,200 to 45,600; a variable coverage rate of 98.7% was achieved; the KL divergence was reduced from 0.138 to 0.025; and the error was corrected to 4.91 gCO2·km-1. According to the above experiment, the hybrid model’s predicted carbon emissions for expressways, arterial roads and local roads are 146.8 gCO2·km-1, 159.3 gCO2·km-1 and 176.7 gCO2·km-1, respectively; at the same time, the root mean square error is 6.48 gCO2·km-1 and the mean absolute error is 5.21 gCO2·km-1. The carbon emission prediction error is less than 5% and thus superior to that of the other model. The proposed hybrid model can address the two problems of poor data quality and difficulty in relationship modelling for carbon emission prediction, thus demonstrating good practicality and engineering application value.
