Reinforcement Learning Enabled Adaptive Test Case Prioritization for Continuous Integration Pipelines

Abstract

Continuous integration pipelines require fast testing, but large test suites can delay feedback after every software change. This study presents a reinforcement learning method for adaptive test case prioritization using recent failure history, execution duration, code coverage, changed-file relevance, dependency information, and test instability. A deep Q-network ranked test cases before each pipeline cycle and updated its decisions from fault-detection rewards and execution-cost penalties. The method was compared with historybased, coverage-based, and random prioritization. Results showed that the proposed approach detected 42% of faults after executing only 10% of the test suite and 63% after 20%, outperforming all baseline methods. Although every approach reached complete detection after full execution, the reinforcement learning method provided much earlier feedback. The framework can reduce testing delay, improve use of limited pipeline resources, and support reliable continuous software delivery in frequently changing development environments.