This article aims to propose a prediction model based on the DART XGBOOST algorithm. Firstly, the DART (with Dropout decision tree) is used for ensemble learning, and finally the XGBOOST algorithm is combined through clustering algorithm. Based on this, MSE and R2 are set as test indicators, and 25 sets of data are used as empirical objects. The results show that: (1) DART XGBOOST has high robustness and generalization performance, Can perform model prediction well (2) When the number of DARTs is 500, the model accuracy converges and reaches the optimal level. This algorithm provides a certain reference for proposing prediction models.