ABSTRACT
The earliest reported attempt at a systematic, evidenced-based approach at detecting the cheating behaviors was focused on probabilistic models identifying collusion and/or answer copying. Psychometrics-based and machine learning-based methods were explored to tackle the complex problem of detecting aberrant testing patterns. Machine learning algorithms are applied to diverse areas such as classification, clustering, regression, and dimensionality reduction. Most of psychometrics-based approaches utilize either item responses or response times or both for identifying aberrant patterns. Person-fit analyses investigate whether the response patterns from test takers deviate from the response patterns from the remaining group under a statistical model of interest. The validity triangle demonstrates the importance and equality of the relationship between psychometrics, content, and security to the validity argument of any test. A number of parametric person-fit indexes have been developed to identify a person's response pattern that is not consistent with the response pattern expected from models based on item response theory.
