The Establishment and Application of Drop-Out-Index of MOOCs Based on XGBoost Feature Selection

Dropout of classes reflects the quality of MOOCs, which is the key issue of online education. In order to predict the dropout rate in advance, this paper presents an efficient prediction framework based on the analysis on real online education data and the prior knowledge of online education. This presented framework combines the feature importance learning and the selection by the classification algorithm of XGBoost, and establishes a Drop-Out-Index (DOI) for online courses. Experiments analysis on massive features extracted from the online-data of XueTang website shows that the feature selection method based on XGBoost achieves better results than other feature selection methods. The validity of DOI has also been verified by testing on different time points in the data set.