Predicting software defectiveness through network analysis

We used a complex network approach to study the evolution of a large software system, Eclipse, with the aim of statistically characterizing software defectiveness along the time. We studied the software networks associated to several releases of the system, focusing our attention specifically on their community structure, modularity and clustering coefficient. We found that the maximum average defect density is related, directly or indirectly, to two different metrics: the number of detected communities inside a software network and the clustering coefficient. These two relationships both follow a power-law distribution which leads to a linear correlation between clustering coefficient and number of communities. These results can be useful to make predictions about the evolution of software systems, especially with respect to their defectiveness.