Multivariate and Naive Bayes Text Classification Approach to Cost Growth Risk in Department of Defense Acquisition Programs.

摘要

In these fiscally austere times, researchers have diligently sought methods to detect cost risk in the DOD acquisition programs. Our research effort reflects a culmination of three years of research seeking solutions to the problem of identifying programs with elevated levels of cost risk. Specifically, we applied multivariate classification and multinomial Naive Bayes text classification techniques to develop three cost risk identification models. We find our model considering a 6-month change in the estimate at complete (EAC) of greater than 5% in magnitude, identified 69.5% of the high- risk programs in our dataset with 76.21% accuracy. Next, our model considering a 6-month increase in the EAC of greater than 5% correctly identified 67.90% of the high-risk programs with 79.68% accuracy. Finally, our model considering a 12-month increase in the EAC of greater than 5%, identified 91.69% of the high- risk programs with an accuracy of 78.31%. This research effort acts as a capstone, concentrating the knowledge collected from previous efforts and provides an actionable decision support tool for the DOD acquisition community. We find this research directly supports the goals of 'more disciplined use of resources' and 'improving efficiency' laid out in the OUSD(Comptroller) FY2013 Defense Budget (Department of Defense, 2012a:3.1).