Identifying Missing Bid Items in Transportation Construction Projects Using K-Nearest Neighbor Model

Missing bid items are an unavoidable challenge in transportation construction projects. Estimators may overlook bid items for various reasons, such as misunderstanding of plans and drawings, miscommunications, lack of experience, and so on. Similarity-based approaches have been widely used in clustering applications but have not been used to identify missing bid items. This research presents the first K-nearest neighbor (KNN) model for proactively identifying missing bid items by identifying and analyzing similar historical projects to the current project. We used the data from 19,865 construction and maintenance projects undertaken by the Texas Department of Transportation from January 2013 to September 2023. We analyzed the impact of different similarity metrics and feature sets on KNN’s predictive capability. Our findings indicate that the Jaccard-based KNN model with intersection over union (IoU) of 66.66% performed best in tests. In contrast, the models using Cosine, Euclidian, and Manhattan distance could not drive improved results by incorporating the quantity of work or unit price information. The increased dimensionality due to considering this information had a detrimental impact on the similarity-based model’s performance, pointing to the need for more complex models to benefit from this information effectively. We expect this model to help practitioners proactively identify missing work items and help researchers develop more advanced solutions.