Reinforcement Learning for Testing Android Applications

Abstract

This paper offers a review of current research studies that use reinforcement learning (RL) to test Android applications. The primary purpose of this study is to simplify future research by collecting and investigating the current state of Android app testing approaches using the RL technique. We provide a well-defined criterion comprising of seven key points. The key points are: addressed problems, reasons for using the RL technique, RL algorithms, supported events, testing techniques, validation, and evaluation methods. In the literature, we have analyzed various techniques to evaluate their efficiency. This study showed that model-based testing is the most commonly used testing technique. Q-learning is the best algorithm in terms of predictive accuracy. We identified that code coverage is the most widely used evaluation metric and comparison with other tools and techniques is the preferred validation approach.