Construction performance classification with controlled information flow in dense engineering activity graphs

Creative Commons License

Mostofi F., Tokdemir O. B., Toğan V.

Journal of Construction Engineering, Management & Innovation (Online), cilt.9, sa.2, ss.1-15, 2026 (ESCI, TRDizin)

Özet

Earned value management (EVM) is critical for monitoring financial and operational performance in large-scale projects. However, its implementation is limited by the complexity of modern construction workflows. Particularly, modeling the cost performance index (CPI) to evaluate the efficiency of activity delivery plans has led to the development of various relational learning models. This study addresses the inability of current relational models to handle high-resolution construction data with dense WBS-based activity networks, where uncontrolled information flow and class imbalance reduce model expressiveness. In this study, we introduce a CPI-specific activity learning mechanism that forecasts CPI using a gated approach to regulate information propagation in dense spatiotemporal data. To contextualize the proposed information modeling framework, this study draws upon a large-scale dataset derived from 77 construction progress reports collected over an 18-month period, which were cleaned and encoded into 52 unique delivery-week categories for model training. We demonstrate that the proposed gated architectures outperform previously proposed attention-based mechanisms. Across eight timeframes of real-world construction datasets, GatedGNN shows superior performance, improving accuracy by 14% and F1 scores by 13% over GCN and GAT models. These improvements enable more reliable CPI classification (efficient vs. inefficient) and support timely corrective decision-making by producing activity-level risk flags that can guide managerial review of resource allocation, schedule coordination, and WBS-specific cost deviations. By improving fidelity in class distribution and managing relational complexity, gated GNNs enable earlier and more accurate identification of inefficiencies. This facilitates proactive decision-making, allowing engineering managers to intervene before performance deviations escalate.