Automated analysis and classification for software issue report using machine learning

Context: Issue reports are used to store the problems that occur using a software and them are submitted by developers and testers. After the software issue is detected then is redirect to an expert that operates in order to solve the problem. This operation is very time consuming, because in this process of creation of the bug report there is possible to find a wrong classification due to human misjudging. Goal: In this thesis’ work, we build a tool, that using machine learning techniques, to classify in an automatic way the issue report with the most probable label class. Method: This works is based over the Mozilla bugs stored in Bugzilla, a bug tracker for general purposes, and it is focused to the correct classification of a new bug. The model works over an implemented version based on the Bugbug project and it creates a classifier with labeled bugs, that can be used with two implementation: OneVsRest or a Binary. Results: Our work is testes over two different scenarios: and a single class behaviour, the class examined is considered as the positive class against all the others, and a multiple classes that groups different classes analyzed individually. The higher result is 60% obtained with the most prevailing cases, while the other class labels obtained an accuracy slightly lower than 30%. Conclusions: Our tool works successful to fulfil the main goal it was designed for. This work can be further expanded in the future with a more homogeneous data set in its classes. It is also possible to improve slightly the precision of the classifier with some minor changes.