Automated Identification of Substantial Changes in Construction Projects of Airport Improvement Program: Machine Learning and Natural Language Processing Comparative Analysis

Contractual changes—mainly substantial changes—within airport improvement program (AIP) projects represent a critical risk that could result in severe negative time and cost impacts. It is critical for airport projects to have in place efficient procedures to process changes effectively, or otherwise this may create an administrative choke point for their stakeholders. Further, with the current US airport infrastructure scoring a D+ (i.e., lacking behind the general US infrastructure), associated authorities called for rebuilding the US airport infrastructure. Thus, it is expected that contractual changes are going to increase for current as well as future US airport projects. This makes it critical to identify these changes early on to incorporate proper change management strategies. However, analysis of contract documents is a process that is known to be inefficient, tedious, and prone to human error. The goal of this research is to create an automated framework to predict substantial contractual changes effectively and efficiently within AIP construction projects. An independent multistep research methodology was used based on principles of natural language processing (NLP) and machine learning techniques (ML). First, the authors adopted a data set containing 876 contractual changes made to the Federal Aviation Administration (FAA) document of guidelines and policies that govern AIP projects (FAA 5100.38D). Second, the authors used NLP techniques to preprocess the aforementioned data. Third, the authors developed hyperparameter-tuned ML models, including k-nearest neighbor (KNN), support vector machine (SVM), random forest (RF), artificial neural network (ANN), extreme gradient boosting (XGBoost), and logistic regression (LR) to predict substantial changes made to the FAA 5100.38D. Accordingly, results indicate that RF showed the most accurate prediction with an area under curve (AUC) value of 0.928, a testing accuracy of 87.45%, and a mean cross-validation accuracy of 92.67%. As such, this automated framework grants stakeholders associated with AIP construction projects a computational decision support tool to easily recognize substantial changes within contract documents, both efficiently and effectively. Ultimately, this research promotes better change management implementation and supports overall AIP project success.