Investment Cost Estimation Method for University Construction Projects Based on Categorical Boosting

University construction projects often lack complete design information during the decision-making and schematic design stages, which makes it difficult to balance efficiency and accuracy in investment cost estimation. To improve the prediction and interpretation of unit area cost, this study develops a CatBoost-based investment cost estimation framework using 72 completed university construction projects. Eleven project indicators related to structural system, foundation type, seismic grade, number of floors, delivery standard, green building rating, fire door type, window type, and finishing configuration were initially considered as candidate input variables. Feature importance was evaluated using RF-MDI, permutation-MDA, and SHAP values derived from the CatBoost model within the repeated five-fold cross-validation framework, and delivery standard was excluded because it showed the lowest comprehensive contribution. The final CatBoost model was compared with RF, XGBoost, GBDT, LightGBM, and BP neural network models using repeated five-fold cross-validation. CatBoost achieved the best average performance among the compared models, with $R^2=0.253\pm 0.229$, RMSE = $696.53\pm 165.41$ CNY/m${}^2$, and MAE = $516.34\pm 112.31$ CNY/m${}^2$. The results suggest that CatBoost provides relatively better prediction accuracy under the current dataset and validation framework, while the relatively low $R^2$ values and performance fluctuations indicate that further validation using larger and more complete datasets is still required.