Application of machine learning methods in test security

Abstract: The post hoc detection of test security has traditionally relied on statistics, but emerging machine learning methods offer enhanced detection performance. To advance the field of test security, we proposed a review of the research literature, categorizing the methods into three major categories: supervised learning, unsupervised learning, and semi-supervised learning. Each of these major categories was further subdivided into three subcategories: ensemble learning, deep learning, and transfer learning. The study elucidated the distinctive attributes of diverse machine learning methodologies, provided practical recommendations for data acquisition and processing, and outlined strategies for input feature selection. Finally, prospective avenues for future research were identified, including machine learning-based person-fit research, machine learning test security research utilizing multimodal data, test security research employing generative adversarial networks, and the interpretability of research results.