Using source code metrics to improve fault localization

Abstract

it lacks the accuracy and provides only limited effort reduction. To overcome the limitations of technique based purely on coverage data, we propose FLUCCs (Fault Localization Using Code and Change Metrics), an approach which extends SBFL technique with code and change metrics, such as size, age and code churn, which have been studied in defect prediction. Using both suspiciousness scores, which are calculated by existing SBFL metrics, and code and change metrics as features, we apply two learning-to-rank algorithms: Genetic Programming (GP) and linear rank Support Vector Machines (SVMs). We evaluate our approach with ten-fold cross validation of method level fault localization, using 210 real world faults from Defects4J repository. As a results, GP with SBFL scores and additional code and change metrics ranks 106 faults at the top and 173 faults within the top 5.

Fault localization aims to support the debugging process by highlighting the program elements which are suspected to be the cause of failure. Spectrum Based Fault Localization (SBFL), which relies only on coverage data of passing and failing test cases, has been widely studied. This, however, has been criticized for being impractical in practice