A Dropout Prediction Framework Combined with Ensemble Feature Selection

In recent years, with the rapid development of large-scale open online courses, low completion rate and high dropout rate have been important challenges for open online courses. Therefore, it is necessary to make effective prediction and timely intervention to ensure the completion of the course. Some of the traditional prediction models only use the features extracted manually from students' clickstream data, which is too subjective to guarantee the quality of features and affect the prediction accuracy. Others generate features automatically with finer granularity, but the problem of feature redundancy appears. In order to solve this problem, this paper proposes a comprehensive dropout prediction framework of MOOCs students. The framework can automatically extract features from clickstream data, and filter features with an integrated feature selection strategy based on clustering and weighted MaxDiff, and finally predict. Experiments show that the model can effectively improve the accuracy of prediction of dropout.