Metrics and Testing Methods for Artificial Intelligence Software Quality Models and Their Application Examples

As artificial intelligence technology rapidly advances, AI software is being increasingly employed across various industries. To ensure the quality and reliability of these software systems, the paper extends the quality measurement and testing methods for artificial intelligence software by utilizing the ISO/IEC 25059 quality model. It not only refines traditional software testing methods to better suit artificial intelligence software but also introduces specific testing methods for quality attributes unique to AI systems, which include five main characteristics: User contrallability ensures timely user intervention in system operations to avoid potential errors or security risks, thereby enhancing system usability; transparency increases the level of information disclosure, allowing users and stakeholders to better understand and trust the AI system, facilitating greater user acceptance; functional adaptability emphasizes the system’ s learning and adaptability in dynamic environments, ensuring it can effectively handle changing data and diverse usage scenarios, which is crucial for maintaining functionality; robustness assesses the stability and reliability of the system under abnormal inputs or attacks, ensuring it can continue operating under all conditions; and intervenability ensures users can quickly and effectively manage system faults, thereby minimizing potential damage. To validate the proposed quality measurement methods, face recognition software based on deep learning was employed as a case study to evaluate its performance through experimental analysis. These validations not only demonstrated the effectiveness of the quality measurement methods in practical applications but also further enriched the testing methods for AI software.