AN IMPROVED MACHINE LEARNING APPROACH TO TEXT MINING FOR AUTOMATIC BUG ASSIGNMENT

Abstract

With the advent of digitization these days, software evolution has taken a deep rise in the computing industries. Large software development projects are being developed which, in turn, have a massive amount of data along with the bugs written in their software repositories. These bug reports are to be managed using a particular bug tracking system, and a diversified group of developers is involved in fixing those bugs. The number of bug report generated on a regular basis by frequently used and accessible systems, are generally high.

To triage the incoming reports manually consumes more amount of time. One aspect of bug triaging is to assign a particular report to the developer with the required proficiency. Automating the process of assigning the bug to the developer with suitable expertise can degrade the software evolution costs and effort.

Previous works have used various machine learning algorithms in order to automate the process of bug assignment but have incorporated limited tools which gave ineffective results with the increased size of the projects. To redress this scenario, this paper employs an improved hybrid machine learning approach, along with the bug tossing graphs to give a graph-based model that can predict the results more accurately. It also gives a comparative analysis of the machine learning algorithms that can be applied and give a solution as to which technique performs better.

The approach used will inevitably suggest developers who have the correct knowledge about dealing with a bug record, based on the identified component obtained from the short description of the bug report. The work begins with examining the impact generated by various machine learning features, including the attributes, classifiers, and the training history. Bug tossing graphs are being used along with the ranked list of developers in order to predict the accuracy of the bug assigned.