Core Concepts
BayesFLo proposes a novel Bayesian fault localization framework that integrates combination hierarchy and heredity principles to address the challenges in fault localization for complex software systems.
Abstract
Software testing is crucial for reliable software development, but existing methods lack a probabilistic approach. BayesFLo introduces a Bayesian model to assess root cause probabilities efficiently, demonstrating its effectiveness over traditional methods in numerical experiments and case studies.
Existing fault localization methods are deterministic and do not provide insights into the probability of root causes. BayesFLo leverages Bayesian modeling to offer a principled statistical approach for assessing root cause risks, integrating structural knowledge for efficient fault localization.
The sheer number of potential root causes poses computational challenges, which BayesFLo addresses by developing new algorithms for efficient computation of posterior probabilities using integer programming and graph representations.
In experiments with different complexities, BayesFLo showcases its superiority over state-of-the-art methods by providing probabilistic risk assessment and informed ranking of potential root causes based on observed test outcomes.
Stats
There are 5 test runs in Experiment 1 with three passed and two failed runs.
In Experiment 2, there are 8 factors with multiple test cases resulting in both passed and failed outcomes.
Experiment 3 involves 8 factors with varying outcomes across different test cases.