A HYBRID MODEL USING CLUSTERING AND REINFORCEMENT LEARNING FOR TEST CASE PRIORITIZATION IN AGILE ENVIRONMENTS

Abstract

Agile development involves rapid and dynamic regression testing to ensure that it keeps pace with the frequent code changes and continuous integration. The traditional methods of test case prioritization (TCP) rely on static historical data and thus cannot handle noisy logs, changing test behavior, and inconsistent failure modes typical of Agile systems. In this paper, the authors describe a hybrid approach to prioritization based on the combination of DBSCAN clustering and Q-learning as the means of overcoming the two challenges, namely the quality of data and adaptability. DBSCAN groups similar test cases and removes noise and Q-learning is used to train an execution order in each cluster using a reward function balancing early fault detection and cost of execution. A mixed dataset of approximately 15,000 test executions of both Defects4J and real CI pipeline are used to test the model. Findings indicate that the given strategy attains an accuracy of 90.5 percent, APFD of 0.87, and a decrease in the overall testing time of 28 percent, which outperform the Random, Greedy, and DeepOrder baselines. It is lightweight and scalable with an easy deployment into CI/CD pipelines, which makes it highly applicable in the contemporary Agile testing process.