LAUSR

Abstract Construction projects delivered using the Design-Build (DB) method include a single contract which defines requirements associated with various disciplines involved in the project. Consequently, DB contractors often need to develop several subcontracts including only a subset of requirements from the main contract. The current manual practices for subcontract drafting are error-prone and time-consuming. The study developed a novel classification model using machine learning for classifying DB requirements into three predefined categories including design, construction, and operation and maintenance. The paper compared various training approaches to perform DB requirement classification including traditional algorithms (i.e., Naive Bayes, support vector machine, logistic regression, decision tree, and k-nearest neighbor), and two state-of-the-art deep neural networks architectures (i.e., convolutional neural network and recurrent neural network). In addition, it examined the effect of different feature types, feature selection, feature weighting, and ensemble methods on the model training performance. The classification models were trained on a large dataset of over 3000 contractual clauses, and the best model achieved an impressive precision of 93.20%, a recall of 93.08%, and an F-score of 92.98%. The study is expected to assist contract administrators in extracting the precise scope of subcontractors in less time and effort.