Mining Software Repositories for Revision Age-Based Co-Change Probability Prediction

Abstract

Changeability is an important aspect of software maintenance and helps in better planning of development and testing resources. Early detection of change-prone entities is beneficial in terms of both time and money and helps to estimate and meet deadlines reliably. Co-change prediction identifies the affected entities when implementing a change in the software system. Recent researches recommend the use of revision history for the identification of co-changed artifacts. However, very few studies are available for investigation of the effect of history size and age on prediction results. This manuscript studies the effect of age of change history on co-change prediction results in software applications by varying the weightage of change commits with time. ROC analysis is done to study the accuracy of the proposed approach, and the results indicate that the older change commits have lower significance in deriving the changeability pattern. The derived change impact set will be useful for software practitioners in change implementation and selective regression testing.