Using Machine Learning to Predict Performance Issues with U.S. Federal Contracts
Open AccessThe U.S. federal government terminated or deobligated over $3.3 billion in awards from fiscal years 2014 to 2020. While it has tried to implement various data analytics programs, the government has not adopted massive machine learning (ML) initiatives across all federal agencies. This praxis asserts that an ML model is needed to predict performance issues that will lead to a reduction in the number of contract terminations for cause. Using fiscal years 2014 to 2020, this praxis supported Hypothesis 1, that ML could be used to predict contract termination. Hypothesis 2, that at least one model would be able to predict performance issues, was also proven; in fact, all the models were useful for predictions. The results indicate that both gradient boosting and random forest models most accurately predict contract termination. Hypothesis 3 was related to which independent variables could predict contract termination. Six variables were most relevant to the ML models: award type, contract pricing, small business, bundled contracts, federal business opportunities, and foreign funded. The federal government can use the six most influential variables identified to avoid performance issues in contracts awarded.This praxis provides a method to predict issues with contracts before they are awarded in the hope that the numbers of terminated contacts and deobligated funds decrease. These results can alert the government to evaluate the elements of a contract more deeply, helping it understand how different variable values can be associated with the likelihood of contract termination. In addition, this praxis enables federal contracting personnel to monitor those contracts that are flagged by the ML models as potential problematic contracts.Keywords: Machine Learning, Federal Procurement, Contract Termination, Gradient Boosting, Random Forest
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.