Abstract: With the rapid development of transportation networks, more and more passengers are choosing air-rail intermodal transportation, which puts forward higher requirements for the recommendation method of air-rail intermodal transit cities. This paper designed a data imbalance handling method that conformed to the characteristics of air-rail intermodal transit city data. The CatBoost algorithm, which can handle categorical features, was used to construct a benchmark model. The model was evaluated on two different test sets with different data distributions, and the accuracy of the model exceeded 85%. Through comparative analysis with other algorithms, it was proven that this model had good stability and better performance, improved the recommendation effect of air-rail intermodal transit cities and better met the travel needs of passengers. Through the analysis of feature contribution, it was found that passenger name characteristics could have an impact on model prediction, which could further improve the personalized recommendation effect of air-rail transit cities.