Predicting Survived and Killed Mutants
Mutation Testing is a powerful technique for evaluating the quality of a test suite. During evaluation a large number of mutants is generated and executed against the test suite. The percentage of killed mutants indicates the strength of the test suite. The main idea behind this is to see if test cases are robust enough to detect mutated code. Mutation Testing is an extremely costly and time-consuming technique since each mutant needs to be executed against the test suite. For this reason, this paper investigates Predictive Mutation Testing (PMT) technique to make Mutation Testing more efficient. PMT constructs a classification model based on the features related to the mutated code and the test suite and uses the model to predict execution results of a mutant without actually executing it. The model predicts if a mutant will be killed or it will survive. This approach has been evaluated on several projects. Two Java projects were used to assess PMT under two application scenarios: cross-project and cross-version. C project was also used to explore if PMT can be applied to a different technology. PMT has been evaluated using only one version of a C project. The experimental results demonstrate that PMT is able to predict execution results of mutants with high accuracy. On Java projects it achieves above 0.90 ROC-AUC values and less than 10% Prediction Error values. On the C project it achieves above 0.90 ROC-AUC value and less than 1% Prediction Error value. Overall, PMT is shown to perform well on different technologies and be robust when dealing with imbalanced data.
Graduation Thesis language
Graduation Thesis type
Master - Computer Science
Dietmar Pfahl, Rudolf Ramler