Overestimation and Underestimation of Software Cost Models: Evaluation by Visualization

Software Cost Estimation (SCE) is a process related to the well-balanced management of a software project. Despite the evolving research activity, the task of estimating accurately the budget and the delivering time has been a research problem for many decades. Nowadays, the cost of a project is still estimated with error. The study of the error produced by estimation models or techniques has been focused on the sources producing it. Usually, the various studies consider underestimations and overestimations of the actual cost to have equal importance. However this is hardly true in practice and such a consideration could be extremely risky for an organization and of course for the customers. In this study, we consider the problem of weighing differently the overestimation and underestimation and we introduce in SCE the utilization of a recently presented graphical methodology, namely the analysis by Regression Receiver Operating Curves (RROC). Our purpose is to evaluate the predictive power of alternative estimation techniques when underestimation and overestimation are not of equal importance. The graphical representation through well-established notions from classification problems provides a straightforward tool for comparing prediction methodologies in different operating conditions. Such a consideration is realistic and desirable for project managers, since underestimation and overestimation of the actual cost have not the same impact on a company. The application of the proposed visualization analysis to real data illustrates the advantages for the critical issues of estimation process.