Authors: K.K. Chaturvedi; V.B. Singh
Addresses: Centre for Agricultural Bioinformatics, Indian Agricultural Statistics Research Institute, Library Avenue, New Delhi, 110012, India ' Department of Computer Science, Delhi College of Arts and Commerce (University of Delhi), Netaji Nagar, New Delhi, 110023, India
Abstract: In the available literature, researchers have proposed and implemented a plethora of bug prediction approaches, which vary in terms of accuracy, complexity and the input data they require, but very few of them has predicted the number of bugs in the software based on the entropy or the complexity of code changes. To use the entropy of code change as a bug predictor, firstly, the history of complexity metric (HCM) defined with different decay weight and decay models were assigned to it (Hassan, 2009). But, they did not propose any method to find out the value of decay rate/factor. In this paper, we proposed a new weight to HCM, a method to find out the value of decay rate/factor and proposed some novel decay-based methods. We have applied simple linear regression (SLR) and support vector regression (SVR) to predict the bugs based on existing and proposed methods of HCM. We have also studied the performance of different complexity of code changes (entropy)-based bug prediction approaches on the basis of various performance measures using four subsystems of Mozilla project. We found that decay models for SVR show better results in comparison with SLR.
Keywords: bug prediction; entropy; software versioning system; software repository; code change complexity; software bugs; simple linear regression; support vector regression; decay weight; decay models; performance measures; software development.
International Journal of Knowledge Engineering and Data Mining, 2013 Vol.2 No.4, pp.266 - 291
Available online: 17 Feb 2014 *Full-text access for editors Access for subscribers Purchase this article Comment on this article