Performance Drop Detector Based on Bayesian Network and Logistic Regression

In this paper, performance drop detector is designed for formative assessment using Bayesian network. Assessment items' difficulty and discrimination are determined by logistic regression to improve detector's performance. Students' result on assessment are predicted by Bayesian network with assessment items as nodes. The probability that the student may fail is calculated based on the properties of nodes for each path, the threshold probability is set as 0.9 to improve the accarucy for the prediction. The precision and recall of the prediction depends on number of layers of the network, the assessment item's difficulty and discrimination. With our sample, the precision of the prediction can be as high as 0.94 (recall 0.30).