Reinforcement Learning Based Test Case Prioritization for Enhancing the Security of Software

Abstract

In order to enhance the security of software, each system update needs to perform regression test. Regression testing in a continuous integration environment requires test cases to meet the needs of rapid feedback. Therefore, it is necessary to enable test cases to be effectively sorted within a certain time range so that more failure data could be discovered and the fault detection rate of testing could be improved. Reinforcement learning algorithms interact with the environment, so it is viable to optimize the sorting problem of test case in the process of continuous integration through a reward mechanism. In the development environment of continuous integration, it has been experimentally proven that the execution history of test cases in the last four cycles has a greater impact on the sorting of test cases in the current cycle. Therefore, a new RHE reward function was put forward by using part of weighted information obtained from historical execution result for enhancing the security of the system. Taking the influence of execution time into account, the multi-target sequencing technology for test case is employed with a view to improving the efficiency of defect discovery. It has been found by applying this sorting method to three industrial testing research that: (1) compared with weighted reward function based on the entire historical execution information, the function based on the four historical execution information had a higher capability of detecting faults; (2) The reward function obtained from the weighted historical results could effectively improve the fault detection rate and reduce the time consumed. (3) The multi-objective sorting methods taking execution time into consideration was able to maximize the number of testing cases that had already discovered faults within the available time.