Poor requirements definition can adversely impact system cost and performance for government acquisition programs. This can be mitigated by ensuring requirements statements are written in a clear and unambiguous manner with high linguistic quality. This paper introduces a statistical model that uses requirements quality factors to predict the probability of satisfying the required system operational performance. This work explores four classification techniques (Logistic Regression, Naïve Bayes Classifier, Support Vector Machine, and K Nearest Neighbor) to develop the predictive model. This model is created using empirical data from current major acquisition programs within the federal government. Operational Requirements Documents and Operational Test Reports are the data sources, respectively, for the system requirements statements and the accompanying operational test results used for model development. A commercial-off-the-shelf requirements quality analysis tool is used to determine the requirements linguistic quality metrics used in the model. The predictive value of the model is confirmed through a sensitivity analysis, cross-validation, and an over-fitting analysis followed by a Receiver Operating Characteristic comparative assessment of model performance. Lastly, five case studies are explored to demonstrate model performance against general data. In all, the results establish that requirements quality is indeed a predictive factor for end system operational performance; and the resulting statistical model can influence requirements development based on likelihood of successful operational performance.
|In Administrative Set: