Interaction among software quality metrics, bug prediction, and test fragility: a machine-learning aided analysis

Context: Software Maintainability is an important and at the same time challenging task, due to its cost and time-consuming factor. One branch of Software Maintainability is bug prediction which in the last decade has attracted much interest in the research community. Goal: The aim of this thesis work is to understand if the ’Bug Prediction’ can be used as a predictor for the ’Test Fragility’, in other words, if there is any sort of correlation between the two. Method: A script has been created that calculates for each project the code quality metrics, uses them to predict the bug-proneness of the classes, and finally calculates the linear regression between the results of the bug prediction and the code fragility metrics. Results: Through linear regression, it was possible to compare the Bug Prediction and the code fragility metrics, it emerged that there is no correlation between the two except in some rare cases. Conclusions: The study demonstrates there is a limited correlation between Bug Prediction and code fragility metrics. However, it must be considered that the analyzed sample was composed only of 30 Android projects. Hence, it would be helpful to repeat the analysis, first of all, on a larger set of projects and then to try on other software families, as well.