A comparative evaluation of decision trees and expert intuition to predict durations in the predesign phase

Abstract: Construction projects are often subject to scheduling errors caused by uncertainty and systematic planning fallacies. This paper evaluates the prediction accuracy of CatBoost and expert intuition to predict the duration of public construction projects in the predesign phase. The authors use a dataset of the city New York (USA) with 367 projects. Both expert intuition and CatBoost are compared with the performance indicator Mean Absolute Error (MAE) and absolute preferrence. The results show high outliers in the expert intuition, while the CatBoost indicates more consistent predictions. From a practical perspective, especially in uncertain events the CatBoost has advantages.