Abstract:The prediction of construction project cost index can effectively solve the cost problems caused by large errors in the preliminary investment estimation of construction projects. Combining the demand for construction cost index prediction models in actual projects, the construction cost index for 2012-2021 released by U city was used as an example to select the optimal prediction model for construction cost index model prediction by comparing the prediction errors between XGBoost and neural network constructed by different feature engineering methods. The results show that the XGBoost model based on tree model feature screening and mean-populated data set has the lowest error in the test set, training set and cross-validation, and can be used as a model for construction cost index prediction.