Towards More Accurate Fault Localization: An Approach Based on Feature Selection Using Branching Execution Probability

The current fault localization techniques for debugging basically depend on the binary execution information which indicates each program statement being executed or not executed by a particular test case. However, this simple information may lose some essential clues such as the branching execution information for fault localization, and therefore restricts localization effectiveness. To alleviate this problem, this paper proposes a novel fault localization approach denoted as FLBF which incorporates the branching execution information in the manner of feature selection. This approach firstly uses branching execution probability to model the behavior of each statement as a feature, then adopts one of the most widely used feature selection method called Fisher score to calculate the relevance between each statement's feature and the failures, and finally outputs the suspicious statements potentially responsible for the failures. The scenario used to demonstrate the utility of FLBF is composed of two standard benchmarks and three real-life UNIX utility programs. The experimental results show that input with branching execution information can improve the performance of current fault localization techniques and FLBF performs more stably and efficiently than other six typical fault localization techniques.