Modeling software component criticality using a machine learning approach

During software development, early identification of critical components is of much practical significance since it facilitates allocation of adequate resources to these components in a timely fashion and thus enhance the quality of the delivered system. The purpose of this paper is to develop a classification model for evaluating the criticality of software components based on their software characteristics. In particular, we employ the radial basis function machine learning approach for model development where our new, innovative algebraic algorithm is used to determine the model parameters. For experiments, we used the USA-NASA metrics database that contains information about measurable features of software systems at the component level. Using our principled modeling methodology, we obtained parsimonious classification models with impressive performance that involve only design metrics available at earlier stage of software development. Further, the classification modeling approach was non-iterative thus avoiding the usual trial-and-error model development process.