Spectral Clustering of Mixed-Type Data for Collusion Detection

• In Event: Test Security and Test Taking Behavior

Abstract

The effectiveness of spectral clustering (SC) on mixed type-data for detecting collusion groups on Computer-Based Tests is explored. Real testing data from a certification exam are used. The level of overlap between the collusion cluster and exact response matches on incorrectly keyed items in a study guide are compared.

Authors

• Soo Ingrisone, HumRRO

• James Ingrisone, Pearson VUE