Events-Based Test Suite Reduction for Mobile App Test Suites Generated by Reinforcement Learning
Abstract
Reinforcement learning is promising for automated test generation. However, a current shortcoming of these algorithms is that the exploration process may result in extra test cases that have duplicate coverage of events within a test suite. Fine tuning parameters of reinforcement algorithms may help, but this comes with trade-offs and requires time consuming and careful consideration of the characteristics of the application under test and its environment. This work takes a different approach. Instead of exploring parameters of reinforcement algorithms, we look at reducing test suites that have already been generated. Specifically, we use test suites that were generated with the SARSA algorithm and then apply a greedy test suite reduction algorithm that uses an event coverage criterion. Results show that test suite reduction results in 10.74% to 50% reduction in test suite sizes while maintaining the same code coverage of the original test suites. The results motivate that redundant event coverage should be considered during and/or after test generation.