An Iterative Group-Based MOPSO with Isomap-Guided Leaders and DQN-Adaptive Parameters for Automated Path Coverage Test Case Generation

Automated Test Case Generation for Path Coverage (ATCG-PC) is a critical yet challenging task in software testing, especially for large-scale programs where metaheuristic algorithms often suffer from premature convergence and inefficient exploration. This paper proposes a novel algorithm, Isomap-DQN-MOPSO (IDMOPSO), which significantly enhances the Multi-Objective Particle Swarm Optimization (MOPSO) framework. Our approach introduces an iterative, prefix-based path grouping strategy to manage complexity. Crucially, it integrates two machine learning-based enhancements: an Isomap manifold learning strategy for more effective leader selection to guide the swarm and escape local optima, and a hybrid Deep Q-Network (DQN) for dynamically adapting learning factors to balance exploration and exploitation. Comprehensive experiments on a diverse set of 18 programs demonstrate that IDMOPSO achieves superior performance, particularly on large-scale programs where it attains significantly higher path coverage rates than state-of-the-art methods. Ablation studies confirm the synergistic effect of combining Isomap and DQN, validating our approach as a robust and scalable solution for complex ATCG-PC problems.