Um método para medição de software com foco na interpretação de medidas.

Abstract

With the evolution of technology and high competitiveness, there is an increasing need for software companies to reduce costs and improve the productivity and the quality of the products delivered. In this context, software measures are essential resources to achieve these objectives. Studies indicate that measures, when used early in the software development cycle, help correct requirements flaws, prevent errors, decrease the development cost, and prevent side effects from late changes. Other studies reveal that, from the data provided by the measurement, the measurement user can view the progress of the projects and make decisions based on objective information. However, despite the benefits described, it is essential to highlight that a poorly designed measurement process can lead to several problems: the collection of irrelevant, redundant, incomplete, or low-quality data, resulting in wasted effort and inconclusive or erroneous data analysis. In this context, to ensure that the measurement user will make the decision based on coherent and representative information of the situation, he/she must understand the value attributed to the measure, considering the factors that influence this value. Thus, given this objective and based on the measurement process covered in ISO/IEC 15939:2007, this thesis’s scope focuses on four steps that impact the interpretation of measures: measure selection, measure validation, data validation, and definition of thresholds or reference values. In the literature, it is possible to find several studies addressing software measures, including ones that cover the steps mentioned earlier. However, most studies, besides not focusing on the interpretation of measures, deal with one or another step in isolation. Based on this, this thesis proposes a method for measuring software, which contemplates the four steps described in an integrated manner, with a focus on supporting the interpretation of measures. Further, it models the uncertainty involved in this problem using Bayesian Networks. The validation of the work was carried out based on a case study and a focus group with industry practitioners. The results show that the method in question is useful for a more confident decision.