Learning the structure of retention data using Bayesian networks
Amy McGovern, Christopher M. Utz, Susan E. Walden, Deborah Trytten · 2008
We introduce a novel approach to examining retention data by learning Bayesian Networks automatically from survey data administered to minority students in the College of Engineering at the University of Oklahoma. Bayesian networks provide a human readable model of correlations in large data sets, which enables researchers to improve their understanding of the data without preconceptions. We compare the results of our learned structures with human expectations and interpretation of the data as well as with cross-validation on the data. The average Area Under the Curve of the networks using cross-validation was 0.6. The domain experts believe the methodology of automatically learning such structures is promising and we are continuing to improve the structure learning process.