An explainable hybrid model of Tourism User Behavior prediction based on SMOTE-XGBoost

This paper proposes an innovative hybrid model based on SMOTE-XGBoost, aiming to enhance the prediction performance of tourist behavior. By analyzing the existing literature, we found that although various machine learning methods have been used for consumer behavior prediction, there are still deficiencies in handling imbalanced data in the tourism sector. We utilized the SMOTE algorithm to perform oversampling and undersampling on the training data, effectively addressing the issue of data imbalance. This approach was combined with the XGBoost algorithm to achieve precise behavior prediction. The model outperforms baseline models across various key performance metrics. Additionally, SHAP analysis revealed that features such as "total likes on out-of-station check-ins received" and "whether the customer follows the company page" significantly influence the prediction of user travel behavior. Through these methods, we not only optimized prediction performance but also provided new strategies and support for enhancing the user experience on tourism websites.