DyEnTRAM- Dynamically Enhanced Metadata based Approach for Bug Assignment

Abstract

Background: In open source repositories, daily numerous bugs are reported and making manual triaging difficult as well as time consuming. Aim: In this paper, we proposed a Dynamically Enhanced Metadata based approach for bug assignment (DyEnTRAM). Unlike TRAM, that worked on additional metadata features which are selected by Feature Selection (FS) algorithms that empirically analyzed the feature value and top value features are selected for triaging. Methods: The work is focused to enhance the bug features as well as rational choice of FS algorithm to improve the accuracy of developer prediction. Approach validation is done on four open- source projects and three machine learning based classifiers namely: Naive-Bayes (NB), SVM and Decision-Tree (C4.5). This approach is only worked with highly ranked meta-data fields which are selected by FS algorithms. Result: This approach has shown better prediction accuracy over the Baseline, X2 and TRAM. Findings have shown F-Score improvements in DyEnTRAM over TRAM approach as much as 24.6%, 10.9%, 16.4%, and 18.2% for Eclipse, Netbean, Firefox and Freedesktop projects respectively. Conclusion: DyEn-TRAM approach effectively improved bug prediction accuracy on the chosen datasets as compared to TRAM, X2, and baseline.