Predicting performance from test scores using backpropagation and counterpropagation

Two neural networks for general mapping problems, backpropagation and counterpropagation, are trained to predict students' grades in Calculus I from placement test responses. The effect of the number of hidden units is investigated. The benefit of including topological structure on the cluster units of a counterpropagation net is illustrated. Noisy data sets are used to train the backpropagation net to improve the ability of the net to generalize.<
>