Weighted Historical Information based Reinforcement Learning Reward for Test Case Prioritization

With the increasing complexity of software systems and the growing demand for efficient testing, reinforcement learning (RL) emerges as a promising approach for test case prioritization. However, existing RL-based methods often underperform due to poorly designed reward functions that overlook historical execution patterns and cost-sensitive factors. To address these issues, this paper proposes a novel reward function which integrates weighted historical information and simultaneously considers both fault risk and execution cost (WHIFR). By dynamically adjusting the reward signal based on historical fault detection effectiveness and execution behavior, the proposed reward function guides the learning agent to make more informed prioritization decisions. Experimental results on multiple open-source datasets demonstrate that the optimized reward function significantly improves fault detection rates and cost-effectiveness compared to existing RL-based methods, thereby validating the effectiveness of the proposed approach.