Response Tabling– A simple and practical complement to Knowledge Tracing

In this paper we introduce a method of predicting student performance by simply calculating the expected outcome of students with the same sequence or subsequence of responses. This expected outcome, which is simply the percent correct, can be calculated for each response subsequence. The combination of expected outcomes for each subsequence can then be combined for a final prediction of a particular student response. Using skill builder problem sets from the ASSISTments Platform we tested this algorithm against an established model of learning called Knowledge Tracing. Both methods utilized the same data which was only student response data. We found that the Tabling method slightly exceeded knowledge tracing in prediction accuracy. The tabling method training time was minimal, taking only a few seconds to train compared to the 30 minute training time of knowledge tracing. We believe this work offers a valuable alternative to knowledge tracing for use with prediction tasks when information about student learning or knowledge is not required.