Exploring Perspectives, Issues, and Practices in the Testing and Quality Validation of AI Software

Artificial intelligence and big data computing are developing at a fast pace, which has resulted in the creation of several software service systems that use different machine learning models. By processing multimedia data for functions including picture identification, recommendation, decision-making, and prediction, these systems seek to facilitate intelligent decision-making in enterprises. On the other hand, there is growing worry about the rising quality problems that are driving up testing expenses for businesses. This study focuses on the quality assessment of AI software functionality features and tackles the dearth of discussion around testing and quality validation for AI software. This study delves into the efficacy of supervised machine learning (ML) algorithms for addressing the test case prioritization problem. Four distinct ML algorithms, including Ranked Support Vector Machines (SVM Rank), K-Nearest Neighbor (KNN), logistic regression (Log Reg), and neural networks, are meticulously evaluated. Additionally, two ensemble learning strategies, historical ensemble learning and combinatorial ensemble learning, are introduced to further enhance prioritization quality. The evaluation methodology involves computing priority values for test cases and assessing effectiveness primarily through the Average Percentage of Faults Detected (APFD) metric. Results indicate that logistic regression excels among individual ML algorithms, while combining all four approaches yields superior performance. The inclusion of test case description features significantly impacts prioritization quality, with logistic regression consistently outperforming other algorithms across different subject systems.