MACHINE LEARNING ALGORITHMS FOR AUTOMATED ASSESSMENT OF THE QUALITIES OF BUILDING INFORMATION MODELS

The article discusses modern approaches to assessing the quality of building information models (BIM) using machine learning algorithms. The main focus is on the application of the CatBoost algorithm for automated calculation of percentage readiness of BIM models based on a list of design comments and errors identified during automatic inspection of construction sites. The advantages and limitations of this method are analyzed and its feasibility in tasks of regression analysis is justified. As part of the study, an experiment was conducted using specialized datasets containing numerical, textual and categorical features that reflect types of errors, criticality and features of individual components of BIM. In addition to the CatBoost model, XGBoost and Random Forest also participated in comparative analysis. According to the test results, CatBoost showed the best indicators: average absolute error less than 0.06 and R2 coefficient ≈ 0.998. Further testing on the new project demonstrated the model's high generalisation ability and resistance to unknown data. Numerical metrics were presented, as well as visualisation of predictions and errors for training and testing sets, confirming the accuracy and absence of systematic bias. A comparative analysis of CatBoost with alternative algorithms was performed based on key criteria: the ability to work with categorical variables, text processing, prediction accuracy, overfitting resistance, and computational resource requirements. The obtained data substantiated the advantages of CatBoost over other gradient boosting and ensemble methods, and confirmed its effectiveness in regression analysis for construction data.