Automatic prediction of bug fixing effort measured by code churn size

Abstract

During software maintenance, developers often receive many bug reports. Project managers often need to manage limited resources to resolve the many bugs that a project receives. To help project managers perform their job, past studies have proposed techniques that predict the amount of time that passes between a bug report being submitted and it being resolved. However, this time period might not be representative of the actual development effort, as developers might not work on the bug right away or all the time. In the open source development setting, developers are only volunteers and might not devote their full working hours to fix a bug in a particular open source project. In the industrial setting, developers might be asked to perform various tasks aside from fixing a particular bug. In this work, we estimate bug fixing effort in terms of code churn size. Code churn size is the number of lines of code that is either added, deleted, or modified to fix the bug. Lines of code has traditionally been used to estimate effort. However, no past studies have proposed techniques to automatically predict code churn size. In this work, using code churn size as estimation for bug fixing effort, we propose a classification-based approach that predicts, given a bug report, whether the bug fixing effort would be high or low. We have evaluated our approach on 1,029 bug reports from hadoop-common and struts2. The result is promising; we can achieve an Area Under the Receiver Operating Curve (AUC) of 0.612 to predict bug fixing effort in terms of lines of code churned, which is a 22.4% improvement over a baseline.