Document Type : Research Paper
Authors
1
Department of Civil Eng., Faculty of Eng., Urmia, Iran
2
Department of Civil Eng., Faculty of Eng., Urmia University, Urmia, Iran
3
Department of Civil Engineering, Ka.C., Islamic Azad University, Karaj, Iran
4
Department of Biosystems Mechanical Engineering, Faculty of Agriculture and Natural Resources, University of Tehran, Karaj, Iran.
Abstract
Rework in large concrete dam projects (LCDP) is a costly and delay-inducing phenomenon, arising from errors, design changes, or execution defects, which severely threatens the quality, productivity, and safety of the structure. This research employs a novel hybrid multi-criteria decision-making (MCDM) and machine learning (ML) approach to identify, weight, and prioritize the factors influencing rework. Data were collected from 26 expert specialists using a two-level questionnaire (10 main criteria and 100 sub-criteria) based on a Likert scale. The MCDM weighting was performed using four methods (Simple Multi-Attribute Rating Technique - SMART, Shannon's Entropy, Criteria Importance Through Inter Criteria Correlation, and Rank Sum) and the geometric mean. This process identified the top three criteria as Design and Technology (weight ≈0.31), Geological and Environmental (0.18), and Legal and Contractual (0.12). The top ten sub-criteria included groundwater infiltration, high cost of digital technologies, cybersecurity issues, defects in concrete compaction, and political pressures (with a low weight gap of ≈0.0008). Among methods, Partial Least Squares Regression (PLSR) demonstrated the best performance with a test mean squared error of 0.0028 and a test R² of 0.9984. Regularized Bayesian regression and Ridge regression models, with R² > 0.997, ranked next. The predicted importance from the top ML models confirmed that geotechnical factors (e.g. groundwater infiltration: ≈3.9), concrete quality (compaction: ≈5.6), and digital technologies (high cost: ≈3.3) were the most critical, showing complete alignment with the MCDM weights. The integrated hybrid approach increased prediction accuracy to 99.8%, providing a dynamic tool for proactive project management.
Keywords
Main Subjects
)