Software Repository Mining Analytics to Estimate Software Component Reliability

Abstract

Finding and fixing software bugs is expensive and has a significant impact in Software development effort. Repositories have hidden predictive information about Software history that can be explored using analytics and machine learning techniques. Current research in Mining Software Repositories (MSR) is capable of ranking and listing faulty components at the file granularity. Our goals are predicting Software defects with method granularity, by mining repositories, and improve current fault localization techniques with the results from defect prediction. We have implemented a tool called Schwa, available for free on Github, that is capable of analyzing Git repositories. For Java and others languages we achieved method and file granularity, respectively. Metrics such as revisions, fixes and authors are tracked and used to feed the prediction model with time relevance. This thesis does the following contributions: a method to parse and represent diffs from patches with method granularity; a model to compute defect probabilities; a framework for mining Software repositories; a technique to learn the importance of tracked metrics; a method to evaluate the gain of using defect probabilities in fault localization.