Abstract: The test security problem refers to the occurrence of abnormal performance of subjects or the leakage of test questions in the process of psychological or educational tests, which seriously affects the accuracy of the test results. To solve this kind of problem, a large number of Person-fit statistic methods have appeared in the field. However, this method has some limitations, mainly in the difficulty of using rich process data, poor fit with empirical data and often based on certain assumptions, while Machine Learning (ML) methods can better overcome the above limitations. This study reviews the research of machine learning in the field of test security from the perspective of different test types and compares the advantages and disadvantages of different machine learning algorithms. Thus, it provides practical applicants and researchers with a reference for the selection of machine learning algorithms and further research ideas.

Key words: Machine learning, Psychological tests, Educational tests, Test security, Person-fit statistic