A Data Driven Approach to the Discovery of Better Cognitive Models

Cognitive models composed of knowledge components are an integral part of intelligent tutors and drive many of the instructional decisions that these systems make. Most of these models are designed by educators and subject experts. Today vast amounts of data, collected from many intelligent tutors, allow us to analyze and improve the current cognitive models through educational data mining. In this research, we show how we identified, in the tutor data, potential improvements to existing cognitive models and then evaluated those improvements using statistical analysis and cross validation.