<p>The advent of the internet age has accelerated the transformation of the Internet economy. The “Internet plus tourism” model has created unlimited development opportunities for online companies and tourism enterprises. However, the vast amount of tourism information on the internet, coupled with the diversity of consumer needs, has made predicting tourism consumption decisions more difficult. To obtain more accurate results for predicting consumption decisions, this research focuses on the tourism decision model of college students on mobile social networks. This model is based on ensemble learning and feature learning methods to build a model structure with two layers of extreme gradient boosting base classifier and one layer of gradient boosting + categorical features classifier. Then, the study employs the greedy method, which uses the maximum node gain feature for splitting, enabling the extreme gradient boosting model to fit quickly. The results indicated that the accuracy of the gradient boosting model with categorical features was 91.71%, with an AUC value of 0.865. The accuracy of the extreme gradient boosting model was 86.27%, with an AUC value of 0.783. The minimum loss value of the two layer integration model (TLIM) based on feature learning and ensemble learning in the training set was 0.225, and the maximum accuracy rate was 91.3%. These findings suggest that the decision model for college students’ tourism consumption based on mobile social networks holds significant reference value in the field of tourism consumption decision-making.</p>