Improving the Quality of Software by Quantifying the Code Change Metric and Predicting the Bugs

Abstract

“When you can measure what you are speaking about and express it in numbers, you know something about it; but when you cannot measure, when you cannot express it in numbers, your knowledge is of a meagre and unsatisfactory kind; it may be the beginning of knowledge, but you have scarcely, in your thoughts, advanced to the stage of science.” LORD WILLIAM KELVIN (1824 – 1907). During the last decade, the quantification of software engineering process has got a pace due to availability of a huge amount of software repositories. These repositories include source code, bug, communication among developers/users, changes in code, etc. Researchers are trying to find out useful information from these repositories for improving the quality of software.

The absence of bugs in the software is a major factor that decides the quality of software. In the available literature, researchers have proposed and implemented a plethora of bug prediction approaches varying in terms of accuracy, complexity and input data. The code change metric based bug prediction is proven to be very useful. In the literature, decay functions have been proposed that decay the complexity of code changes over a period of time in either exponential or linear fashion but they do not fit in open source software development paradigm because in open source software development paradigm, the development team is geographical dispersed and there is an irregular fluctuation in the code changes and bug detection/fixing process. The complexity of code changes reduces over a period of time that may be less than exponential or more than linear. This paper presents the method that quantifies the code change metric and also proposed decay functions that capture the variability in the decay curves represented the complexity of code changes. The proposed decay functions model the complexity of code changes which reduces over a period of time and follows different types of decay curves. We have collected the source code change data of Mozilla components and applied simple linear regression (SLR) and support vector regression (SVR) techniques to validate the proposed method and predict the bugs yet to come in future based on the current year complexity of code changes (entropy). The performance of proposed models has been compared using different performance criteria namely R2, Adjusted R2, Variation and Root Mean Squared Prediction Error (RMSPE).