Fault Localization with Partially Reliable Test Results Using Dempster-Shafer Theory

Fault localization is a critical procedure in software development process. Previous studies based their research on the precondition that test results are conveniently acquired and 100% correct, which does not happen in the real world. In this article, we propose the concept of γ- reliable test-suite to demonstrate the potential unreliability of test results. By modeling this unreliability using Dempster-Shafer theory, we managed to pin down the faults under the new situation. Experiments were conducted on both 100%-reliable and partial-reliable Siemens Test Suite and compared against several known spectrum-based localization algorithms, namely Naish1, Naish2, Binary, Wong1 and Russel&Rao. The results proved the prior performance of our approach. We conclude that fault localization problem is more precisely modeled by Dempster-Shafer theory than common statistical theory.