Using predictive models to evaluate the quality of a test suite at class and method level.

Abstract

Testing is an indispensable part of the software development process and is a continuous process during the development life cycle. In this context, examining the behavior of software systems to reveal potential problems is a crucial task. To this end, the test suites usually are utilized to examine the software quality. However, test suite quality control is hard for the tester, especially in an evolving system. Such control is needed to assure and improve the test suite's quality and the application as a consequence. Currently, test coverage criteria are used as a mechanism to assist the tester in analyzing the test suite (e.g., find the weaknesses, and add a new test case or test inputs). However, more strong coverage criteria (potentially showing less glaring weaknesses) are challenging to assess. In this work, we propose a different approach to support the developer in evaluating the test suite quality based on more powerful test coverage criteria. We will follow the Knowledge Discovery in Database process using machine learning algorithms to estimate the prime path coverage at the method and class level. For this purpose, we will create two large datasets consisting of source code metrics and test case metrics from 12 open-source Java projects, and these datasets will be used in the training process to build the predictive models. Using the built models, we expected to predict the prime path coverage at the method and class level with a reliable prediction performance.