A Novel Learning Early-Warning Model Based on Knowledge Points and Question Types

Learning early-warning is of great importance to many educational domains, such as adaptive learning and personalized teaching, and has drawn numerous research attention in recent decades. In order to solve the problem of large prediction granularity in previous study. In this study, we seek to construct two novel features, including knowledge points and question types, and predict students' performance based on the two types of information. According to the predicted results, we divide the early-warning into 3 levels, and provide different levels of guidance and reminders for different warning levels of students. We did experiments based on the two types data of 141 students in Peking University. The result shows that our method has been significantly improved compared with Linear regression, RF and Adaboost. The experiment shows that the model's predicted grades and the real data Pearson correlation coefficient is 0.890568, and the accuracy of predicting warning levels is 85.81%.