Amir Elmishali, Roni Stern, Meir Kalech
AAAI 2016: 4003-4009
Software fault prediction algorithms predict which softwarecomponents is likely to contain faults using machine learningtechniques. Software diagnosis algorithm identify the faultysoftware components that caused a failure using model-basedor spectrum based approaches. We show how software faultprediction algorithms can be used to improve software diagnosis.The resulting data-augmented diagnosis algorithmovercomes key problems in software diagnosis algorithms:ranking diagnoses and distinguishing between diagnoses withhigh probability and low probability. We demonstrate the ef-ficiency of the proposed approach empirically on three opensources domains, showing significant increase in accuracy ofdiagnosis and efficiency of troubleshooting. These encouragingresults suggests broader use of data-driven methods tocomplement and improve existing model-based methods.