Abstract—Continuous Integration (CI) pipelines require fast regression feedback under strict time budgets, making full regression suites impractical for large systems. Deep Adaptive Regression Test Sequencer (DeepARTS) generates an ordered regression test sequence per CI cycle under a budget K. The approach combines a Graph Attention Network (GAT) for code–test dependency encoding, a Transformer encoder for context-aware modeling of candidate tests, and a deep Q-network for sequential test selection using a detection-oriented reward with an execution mask. Continual learning based on entropy-guided replay and Maximum Mean Discrepancy (MMD) drift detection updates model parameters across cycles without full retraining. Evaluation uses Regression Test Prioritization Torrent (RTPTorrent), a dataset containing 20 open-source Java projects and more than 100,000 TravisCI build logs, with chronological per-project splits to reduce temporal leakage. Under identical candidate sets and budget \mathbit{K}, DeepARTS improves early fault detection compared with Linkage Learning-based Non-Dominated Sorting Genetic Algorithm (L2-NSGA), achieving Average Percentage of Faults Detected (APFD) = 0.94 with Precision@K = 99.2% and Recall@K = 98.7%, while maintaining per-cycle recommendation latency near 2.5 s on Central Processing Unit (CPU). Generalization beyond Java projects and TravisCI logs remains to be validated.
 
Keywords—adaptive regression test sequence, optimum regression testing, machine learning, adaptive test sequence recommendation system


Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).