Design of an iterative adaptive method for volatility-aware test case prioritization in rapidly evolving software...

Area of Science:

• Software Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Test Case Prioritization (TCP) is crucial for efficient software testing.
• Traditional TCP methods struggle with dynamic software changes and volatility.
• Optimizing test execution order is key to reducing costs and improving fault detection.

Purpose of the Study:

• To propose an adaptive, deep reinforcement learning-driven framework for volatility-aware Test Case Prioritization (TCP).
• To enhance the effectiveness and efficiency of TCP in dynamic software environments.
• To provide a dependable and understandable TCP solution balancing multiple objectives.

Main Methods:

• Developed a five-module framework integrating Dual-Attention Temporal Graph Prioritization Network (DAT-GPN), Reinforcement-Driven Volatility-Aware Clustered Prioritizer (RD-VACP), Uncertainty-Regularized Multi-Agent PPO Scheduler (UR-MAPPO), Counterfactual Impact Analysis Prioritizer (CIAP), and Multi-Objective Adaptive Ensemble Prioritization Framework (MO-AEPF).
• Employed temporal and contextual attention mechanisms, Q-learning, multi-agent PPO with uncertainty regularization, structural causal inference, and ensemble learning.
• Utilized historical execution logs and software modification data for dynamic graph modeling.

Main Results:

• The framework effectively addresses volatility-aware optimization for improved TCP.
• DAT-GPN assigns priority scores using historical data and dynamic graph analysis.
• RD-VACP optimizes execution order and clusters test cases based on volatility.
• UR-MAPPO enhances policy stability in dynamic scenarios using uncertainty.
• CIAP enables risk-aware decision-making through counterfactual analysis.
• MO-AEPF balances detection time, risk, and resource consumption.

Conclusions:

• The proposed adaptive framework offers a robust and interpretable solution for Test Case Prioritization.
• Integration of reinforcement, causal, and sequential learning provides risk-sensitive, optimal execution.
• Multi-objective ensemble optimization ensures resource efficiency and balanced fault detection.