Artificial Intelligence for Project Managers: Predicting Construction Delays

Abstract

Delays in construction projects are among the main causes of cost overruns and loss of confidence in the sector, especially affecting micro, small, and medium-sized enterprises (MSMEs) that face fragile planning conditions and limited access to predictive analytic tools. This study proposes a methodology based on the CRISP-DM framework to anticipate delays using supervised machine learning techniques, such as Random Forest, Support Vector Machines, and Artificial Neural Networks, along with unsupervised methods like hierarchical clustering. A synthetic dataset simulating construction project conditions was used, and different strategies were evaluated to define thresholds for delay classification. The results show that hybrid approaches outperform fixed threshold definitions, achieving high accuracy (97%) and revealing latent patterns, such as “big delayed little projects,” which highlight the vulnerability of small-scale works to schedule overruns. The developed methodology demonstrates that, with just a few simple variables like built area, project duration, and cost per square meter, MSME project managers can implement replicable predictive models that improve planning and decision-making.