Finding Your Way, Courtesy of Machine Learning

In this paper, we consider the multi-label, multi-class prediction problem in the context of classifying undergraduate education requirements at Stanford University – namely, given a course’s description, determining what general education requirements it satisfies. Given the extremely tough challenge of using a small dataset, we opt to consider models and make modifications that are able to handle multiple classes and produce multiple labels with very limited training examples. We present a slightly modified version of Naive Bayes, an application of BoosTexter, and several iterations of linear classification techniques. Then, we explore the performance of techniques from deep learning including RNN, LSTM, and GRU.