Uma abordagem para a definição de valores de referência de métricas de software baseada em contexto usando redes bayesianas.

Abstract

Metrics play a key role in the software quality management process. Despite their potential benefits, they are generally only used for quantification, not providing adequate support to the decision-making process during the software’s life cycle. To enhance the use of metrics, it is essential to define meaningful reference values (i.e., thresholds), thus giving meaning to the data collected. This work aims to propose an approach to define the software metrics’ reference values according to the project’s context. The approach consists of using the specialists’ elicited knowledge to define context factors that influence the metric’s reference values. This information is used to build a Bayesian network that can be used to aid in the decision-making process. The proposed solution was evaluated through a pilot study conducted with three managers of real software development projecst. Data were collected from the software project managers in order to build Bayesian networks to identify and validate reference values for the Number of Minor Bugs, the Number of Static Analysis Alerts, and the Code Coverage Percentage metrics. Each metric was validated in three scenarios. The proposed approach has shown to be promising in helping professionals to identify representative reference values, promoting a more assertive decision making when it comes to the software project management process