Mutation Test Optimization Using Machine Learning Technique

Mutation testing is a fault-based technique used in software testing.It uses a set of artificual faults(or mutants),to simulate the faulty execution of the program,and measure the test suite adequacy.In practice,engineers tend to use mutants which are difficult to be detected by structural coverage testing.These mutants,called stubborn mutants,can guide test generation,as well as improve the quality of test suites.The paper proposes a technique for mutation test optimization using macine learning.It uses the program context around the faulty statement as features,to predict stubborn mutants by applying machine learning approach.More specifically,this paper:(1)defines fault detection context model,which is a graphic model centered on the faulty statements,and describes the code locations correlated with fault detection via program dependence.(2)It proposes the fault detection document model,which traslates the fault detection context to the words seqeuence,so to simplify the proceeding of learning model.(3)It developes an attention based representation learning model for generating the feature vectors of each mutant,which are used as the direct input to the learning model.(4)It applies a set of learning models,such as decision tree,logistic regression and neural networks,to predict the stubborn mutants within the program under test.Finally,the paper performed an empirical analysis over 26 C programs with 108,000 mutants to evaluate the effectiveness of the proposed method.The results show that,after using the fault detection context as features our technique is more effective in identifying the stubborn mutant.Meanwhile,the machine learning technique can achieve a high precision and recall in predicting the stubborn mutants,which validate the proposed technique in this paper.