Research On The Prediction Method Of Dropout Rate In MOOCs

In recent years,with the rapid development of information technology,large-scale online education,as an emerging education method,continues to flourish.Among them,online learning represented by MOOC is gradually entering thousands of households.Although MOOC is very popular among people,it still faces a huge challenge-high dropout rate,which has affected its further development to a certain extent.Predicting the dropout rate in advance can take measures to avoid as many dropouts as possible.So how to analyze and accurately predict the dropout rate in the massive behavior data of learners is a problem that needs to be solved.Although there are many researches on learner behavior analysis and dropout rate prediction on the online platform,there are still the following problems:(1)The analysis of learners’ behavior failed to combine psychological cognition to reasonably quantify the behavior.At the same time,there was no deep processing of complex time-series data in data processing,and it was impossible to mine hidden behavioral characteristics of learners.(2)When mining potential groups for learner behavior clustering,clustering is based only on the learner’s learning behavior level,and the focus is too much on the behavior and whether the behavior occurs.Meanwhile,they did not proceed from the level of course selection,ignoring its initial preferences and learning goals.(3)The lack of interpretability in the prediction of dropout rate,most of the current research regards it as a traditional binary classification problem or a time series problem,neglecting not only the sequence of events but also the impact of historical behavior and learning that affect learners’ behavior.Meanwhile,when learners just joined the new course,we lacked behavioral data and could not make predictions.In order to solve the current problems,the research content and contributions of this article are as follows:(1)In-depth analysis of each learning behavior,exploring the correlation between each behavior and dropout rate,and modeling and quantifying learner behavior based on psychological background knowledge,so that it can effectively distinguish different learners,and find out their interest in resources and their preference for behavior.(2)Perform time-series data processing on the learner’s behavior in the course to encode and reduce dimensions,extract the learner’s potential behavioral features from it,and then explore the course selection behavior characteristics of the learner from the perspective of course selection.Finally,combining the learning behavior and course selection behavior to effectively tap the potential groups..(3)In view of the deficiencies in the dropout rate prediction method,we propose a deep model.We predict the dropout rate based on the deep learning architecture of the attention mechanism of the association cycle,and use historical behavior as the predictive factor through the attention mechanism of the association cycle.Then innovatively combine the impact of time-series behavior and historical behavior to make predictions.Finally,the potential groups obtained based on clustering are used to predict whether the learner will drop out of the new course just joined,you can predict in advance and adopt relevant measures to intervene as soon as possible.Finally,the rationality of the potential groups mined and the effectiveness of the proposed dropout rate prediction method are verified on different data sets.At the same time,the potential groups mined are used to predict whether the learner will drop out in the new course.The results illustrate that it is feasible to use group behavioral characteristics to predict whether individuals will drop out in the new curriculum.