Evaluating Machine Learning Predictions for Budget Overruns in U.S. Mass Transit Projects: A Correlational Study
Evaluating Machine Learning Predictions for Budget Overruns in U.S. Mass Transit Projects: A Correlational Study
Loading...
Issue Date
2025-10
Authors
Boudreau, Paul
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This study examined supervised and unsupervised machine learning methods to determine if statistical correlations could be established to validate mass transit project budgets. Capital budget forecasts for new mass transit projects in the United States have consistently been underestimated. These budget overruns resulted in funding allocation challenges for the Federal Transit Administration and obligated local transit authorities to secure additional financing. Developing accurate budget estimates is a significant challenge for project managers, particularly given the influence of cognitive biases such as optimism bias. Data from 108 projects were collected from publicly available U.S. government documents to serve as input for the machine learning models. Using a neural network and k-means clustering in SPSS software, the study created models that explained 94.6% (R2 = 0.946) and 88.2% (R2 = 0.882) of the variance in project budget outcomes. The findings highlighted the importance for organizations to apply AI-based machine learning technology to validate budget forecasts. This research built on prior research by applying machine learning methods that had been explored in other project contexts and problem domains. By employing SPSS software with configurable settings, rather than relying on customized Python code, the study enhanced the transparency and accessibility of machine learning regression analysis. The results demonstrated that a budget validation process could be developed using reproducible, data-driven predictive models that incorporate both project-specific and environmental variables.
Description
Keywords
AI Project Budget Mass Transit , Artificial Intelligence , Business, Engineering, Science, & Technological Innovation