Deep Reinforcement Learning based Android Application GUI Testing
Eliane Collins, Arilo Dias Neto, Auri Marcelo Rizzo Vincenzi, José Alberto Maldonado
Abstract
The advances in mobile computing and the market demand for new products which meet an increasingly public represent the importance to assure the quality of mobile applications. In this context, automated GUI testing has become highlighted in research. However, studies indicate that there are still limitations to achieve a large number of possible combinations of operations, transitions, functionality coverage, and failures reproduction. In this paper, a Deep Q-Network-based android application GUI testing tool (DeepGUIT) is proposed to test case generation for android mobile apps, guiding the exploration by code coverage value and new activities. The tool was evaluated with 15 open-source mobile applications. The obtained results showed higher code coverage than the state-of-the-art tools Monkey (61% average higher) and Q-testing (47% average higher), in addition, a greater number of failures.