Harnessing machine learning and structural equation modelling to quantify the cost impact of rework in bridge projects

• Published in: Asian journal of civil engineering. Building and housing Vol. 25; no. 5; pp. 3929 - 3941
• Main Authors: Elseufy, Sara M., Hussein, Ayman, Badawy, Mohamed
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.07.2024
• Subjects: Building Materials; Civil Engineering; Engineering; Sustainable Architecture/Green Buildings; Rework; ANN; KNN; SVM; Structural equation modeling; Machine learning
• ISSN: 1563-0854; 2522-011X
• DOI: 10.1007/s42107-024-01021-z

The study addresses the significant challenge of rework in the construction industry by leveraging machine learning techniques. Specifically, the aim is to develop models that accurately classify the impact of rework causes on the cost performance of bridge projects using objective data sources. Pertinent rework sources and determinants were identified, and a multivariate dataset of prior projects’ cost performance was assembled. Additionally, a structural equation model was developed to calculate the impact of these factors on cost performance in bridge projects. To create a suitable dataset for machine learning, 272 responses from subject matter experts were utilized. The study explores Ensemble techniques, K-Nearest Neighbors (KNN), Artificial Neural Networks (ANN), and Support Vector Machines (SVM). Cross-validation tests were conducted to assess the predictive abilities of the models, and the evaluation results indicated that the SVM model provides superior predictive performance for the dataset examined. SVM achieves 98.53% (89.54%) accuracy in training (testing) with a 1.47% (10.46%) misclassification error. Comparisons were made regarding the impact of rework on cost, with SVM achieving the highest recognition rate across all data divisions, followed by ANN. Conversely, KNN exhibited the lowest recognition rate among the classifiers. With a maximum recognition rate of 97%, SVM emerged as the best classifier. The optimal data separation for testing and training data was determined to be 10% and 90%, respectively. Overall, this study harnesses the power of machine learning to facilitate evidence-based decision-making, enabling proactive prediction of the impact of rework on cost performance in bridge projects.