The Effect of Model Granularity on Student Performance Prediction Using Bayesian Networks