Spectrum Impact Analysis of Fault Proneness Statement for Improved Fault Localization.

Background: Fault localization is an important approach aimed at discovering faults in source codes to accelerate the activities of software development and maintenance. Spectrum-based fault localization (SBFL) techniques have been widely used to find faults. Ranking the fault-proneness statement is a promising research topic and several methods have been proposed in research in recent decades. Aim: To facilitate and interpret ranked files by software quality teams, we thoroughly explore and examine the effectiveness and importance of SBFL in locating faults by utilizing statement-hit spectra. Method: We conduct an empirical study for classifying fault-prone statements by adopting ranking-based fault localization approaches. We set up an experimental environment named Classification of Fault Proneness Statement (CFPS) that automatically classifies fault-prone statements (from very high to low). We conduct an extensive set of experiments on real-world dataset, which are coded and complied in C programming language, to validate the proposed methodology. The experiments are compared and evaluated with four state-of-the-arts similarity coefficient ranking algorithms. Results: Our experiment reveals that the similarity coefficient, tarantula significantly outperforms the others when validated with CFPS. Tarantula with CFPS achieves the score of Mean Average Precision (MAP) and Mean Reciprocal Rank (MRR) 33.27% and 32.91%, respectively on average across the selected experimented programs. Conclusions: The empirical study demonstrates the positive impact and effectiveness of SBFL and CFPS in classifying fault-prone statements. Furthermore, CFPS provides extra and essential information to developers for accurate fault localization and should be considered by software quality teams. The data and codes are released at https://github.com/sagarwhu/SBFL.