Methodical Mapping Research on The Incorporation of Machine Learning into Automated Test Generation

Effective automated test generation may be made possible by machine learning (ML). We highlight new research by examining methods for testing, researcher plans, ML approaches used, assessment, which is and difficulties in this particular field. We use a sample of 124 papers to do comprehensive mapping research. ML either boosts the performance of existing iteration methods or generates input for system, GUI, unit, performance, and combinatorial testing. Additionally, test verdicts, property-based, and expected outputs oracles emerge using machine learning. Along with to the ubiquitous use of supervised learning, which is frequently based on neural networks, and reinforcement programming, which is frequently based on Q-learning, some sources also use unsupervised or semi-supervised learning. (Un-/Semi-) While reinforcement learning is often evaluated using assessment metrics connected to the reward function, supervised techniques are assessed using typical testing metrics and ML-related metrics (e.g., accuracy). The findings we obtained can be used as an outline and point of design ideas for study participants in this area, but there are still lingering problems with training data, retraining, scalability, evaluation complexity, ML algorithms used—and how they are used—benchmarks, and the ability to be replicate.