SBFSelector

Abstract

Tracking changes in code using revision history shared by collaborative teams during software evolution improves traceability. Existing techniques provides incomplete and inaccurate revision history due to lack in detection of renaming and shifting at file, class, and method granularities simultaneously. This research analyzes and prioritizes the metrics responsible for detecting such changes and update the revision history. This improves the traceability by tracking complete and accurate revision history that further improves the processes related to mining software repositories. It proposes SBFSelector algorithm that uses Jaccard Similarity and cosine similarity based on the prioritized metrics to identify these changes. Result shows that 73% metrics belongs to size and complexity that holds more significance over remaining categories. Random forest is best classifier for tracking changes with 0.99 true positive rate and 0.01 false positive rate. It improves traceability by increasing the Kappa statistic and true positive rate as compared to Understand tool.