Research On MOOC Students' Dropout Prediction Model

Today,the online courses have attracted more and more students because of its accessibility and the richness of resources,among which MOOC is a rising star,the amount of its resource is huge and there are much excellent resource of famous universities,so it is widespread in the world quickly.However,the MOOC's constraints for students are so weak that many students terminate the learning course for some internal or external factors,resulting in a waste of educational resources.In order to reduce the occurrence of this phenomenon,we hope to establish a MOOC student dropout prediction model,from which we can find students' tendencies to dropout of school in time,and then give some appropriate intervention.This paper studies the dropout model from two kinds of models.The first kind is the temporal model in which our work concentrates on the RNN model,because the student's learning process has a very obvious time sequence and the students' last time performance may have an influence on the next time performance.We also apply the Deep RNN model to the dropout prediction at the first time.The second kind is part of the discriminative models.The discriminative model makes our feature engineering methods more suitable because they are more extensible,while the temporal model is more suitable for the time-related features.At the same time we have utilized some of the new features not mentioned in the previous paper to improve our model performance,including our proposed Dependent Features(DF),which have significantly improved our model performance.Finally,to mitigate the effect of data imbalance problem on the dropout prediction model,we analyze the effect of resampling method and adjustment for class weight on the dropout model.The results of this study show that the temporal model is lacking in running performance and expansibility of features,and the effect of the feature engineering method on the final result is very obvious.Among the new features we have found the dependent features are very good features,with which the performance of the model has been significantly improved.