Predicting the cost of rework in high-rise buildings using graph convolutional networks

To reduce the risk of unexpected cost of rework (COR), a variety of predictive models have been developed in the construction management literature. However, they primarily focus on prediction accuracy, and rather less attention has been paid to the trustworthiness of prediction models. This increases operational risk and hinders its integration in related decision-making. Aiming to reduce the utilization risk and increase the reliability of COR prediction models, this study exploits the graph convolutional network (GCN) model, which enhances representativeness by accommodating interrelationships among the root causes of nonconformances. The GCN can process a more representative input network that provides COR records while factoring in the shared root causes of nonconformance in the resulting COR. The proposed approach achieved a COR prediction accuracy as high as 85%, which is significantly higher than that of any existing cost prediction model. The demonstrated accuracy and lower risk of the proposed GCN model thus enhance the reliability of the prediction and trust in its outcome, facilitating its integration into developing rework prevention strategies and relevant resource allocation for construction professionals. The study contributes to construction project management by proposing a novel COR prediction model that embodies accuracy, representativeness, and interpretability. Whereas we tailored the GCN model to predict COR with a focus on nonconformance root causes, it is noted that rework costs can also be influenced by other project factors, such as site safety.