Predição de falhas em projetos de software livre baseadas em métricas de redes sociais

Abstract

This work was developed with the goal of analyze human factors (communication and cooperation) as parameters for failure prediction, seeing that problems with communication and cooperation can lead to failure of projects. Many works of software engineering analyze communication and cooperation among the developers, by extraction data from control version systems and of communication mechanisms, among then, some use social networks and SNA to analyze the communication and the cooperation among the developers. The experiment realized in this work, consisted in the generation of matrices (developers vs developers) based on data of communication and cooperation extracted from the repository of open source software, this matrices represents social networks of the developers, then was calculated the values of metrics of SNA, and this metrics were utilized in the training set and validation set of Bayesian classifiers that actually realizes the prediction and lastly was used the statistical measures Recall/Precision/F-measure to evaluated the performance of the predictions. From the experiment realized was possible verify that cooperation can be used to failure predict, and that communication may not get a satisfactory performance to failure prediction.