USING THE EM ALGORITHM TO PREDICT 1ST YEAR UNIVERSITY PERFORMANCE

Abstract

The problem of predicting first year university performance on the basis of high school results alone is considered. The method investigated uses the EM algorithm to estimate the mean mark for each subject and the covariance for each pair of subjects that would have arisen, had it been possible for all the students to take all the subjects both at high school and at university. A conditional distribution can then be obtained for any set of proposed university units given the high school marks of the any particular student. This method (NEM) is contrasted with several more traditional methods—linear and logistic regressions—designed specifically for particular subjects and courses. The NEM me..