Abstract: With the prosperity of high-speed railway station media advertising market, a scientific, systematic and comprehensive evaluation system of high-speed railway station media resource value is urgently needed to guide the management of media resources. This paper studied the relationship between the multi-dimensional data of the value evaluation index system and the value of high-speed railway station media resources, with the help of feature engineering, extracted core data features that were strongly related to the target. Using a variety of regression learning algorithms, the paper selected the eXtreme Gradient Boosting (XGBoost) algorithm with the best evaluation index, and constructed the value evaluation model of high-speed railway station media resources. Through model optimization, the goodness of fit value was improved to reach the target value of 0.8. The application proves that the deviation degree of the model does not exceed 15%, which can provide reference for the daily operation and pricing decision of high-speed railway station media resources.