Uma abordagem baseada em redes bayesianas para auxiliar a interpretação de métricas de software.

Abstract

Despite the large amount of software metrics that has been proposed since the 1960s, their adoption and application is still limited in many situations. A challenge in using metrics is to interpret them to make assessments and predictions regarding software development projects. Several researchers proposed approaches to define thresholds to determine whether the value of a metric is acceptable, in order to help the developers and managers to interpret it. These approaches, however, do not consider risks and other subjective factors that have impact in the measurement process and might influence the metrics’ interpretation and consequently the manager’s decision. Other researchers proposed models combining software metrics and subjective factors to assist on decision-making, but they did not consider interpretation risks such as problems in metrics’ collection and reporting process and metrics misuse. In this research, we propose an approach to construct Bayesian networks to assist on metrics interpretation considering these risks. The Bayesian networks constructed help the managers to identify risks related to the metrics and controller factors to mitigate them. The goal is to maximize the metrics’ accuracy and minimize wrong decisions based on software metrics. The approach was successfully validated with a case study performed with four projects and we concluded that it’s a promising approach to assist practitioners to interpret metrics and support software projects managerial decision-making.