Automatic Equivalent Mutants Classification Using Abstract Syntax Tree Neural Networks

Abstract

Mutation testing is a testing technique that is effective at designing tests and evaluating an existing test suite. Even though mutation testing has been developed to be applicable and effective towards different types of software systems and programing languages for many years, wide industrial use of mutation testing has not yet been seen. One primary reason that prevents developers and testers from using mutation testing is the expensive computational cost. Specifically, the need to manually identify equivalent mutants is a major obstacle and makes mutation testing very time consuming and labor intensive. This paper addresses this limitation and proposes a machine learning-based approach that designs and trains an abstract syntax tree recurrent neural network model to automatically classify equivalent mutants during the process of mutation testing. A pilot study with 582 mutants shows that the proposed machine learning-based approach can automatically classify equivalent mutants with an accuracy higher than 90%. The approach can significantly save the manual effort and time spent on identifying equivalent mutants during the process of mutation testing.