Research On The Prediction Model Of MOOC Dropout Rate Based On Deep Neural Network

With the popularization of "Internet+education",the development of online education has spawned many new educational patterns and concepts.Since the rising of MOOCs(massive open online courses)in 2012,it has rapidly developed globally by virtue of the technical support of education big data and artificial intelligence.MOOCs attracts tens of millions of learners from all over the world with its rich curriculum resources and high-quality course related services,but the high dropout rates seriously restricts the further development of MOOCs and becomes the most urgent problem to be solved.Learners' learning behaviors are made full use to construct an effective MOOCs dropout rates prediction model,which can provide managers with decision-making basis and alleviate the occurrence of such phenomena.Therefore,the key issue MOOCs dropout rates prediction is studied in this paper.In recent years,although Chinese and foreign scholars have carried out model construction and theoretical research on the prediction of MOOCs dropout rates,there are still many deficiencies to be solved.Traditional prediction models of MOOCs dropout rates mainly adopt the manual methods to extract features from clickstream data,which is time-consuming and laborious and ignores the impact of learners' information and course information on the dropout rates of MOOCs.Deep learning has a very strong feature expression ability and can automatically extract features from original data,so it is widely used in various fields.Therefore,a prediction model of MOOCs dropout rates based on deep neural network is constructed by using the theories related to deep learning,so as to put forward new ideas for improving the prediction performance of MOOCs learners' dropout rates.First of all,this paper reveals the latest research status and trends of predicting MOOCs and MOOCs dropout rates through constructing the mapping knowledge domain of MOOCs research hotspots,the authors of MOOCs research and the mapping knowledge domain of the institution.Meanwhile,ANOVA analysis method and Apriori algorithm are adopted to analyze the correlation between learners' information,learning behavior and MOOCs courses,so as to explore the impact of learners' information and course information on MOOCs dropout rates,summarize the reasons for MOOCs learners'dropout,and provide a theoretical basis for the construction of the prediction model of MOOCs dropout rates.Secondly,aiming at the disadvantages that traditional MOOCs dropout rates prediction methods rely on manually extracting features by feature engineering,the extraction process is time-consuming and laborious,and the feature extraction strategy is not universal,this paper proposes a CNNGRU MOOCs dropout rates prediction model by combining convolutional neural network with GRU network.The model has the ability to automatically extract local features from the original MOOCs data with the help of the convolutional neural network.At the same time,since learners' learning behavior is related to time series,the model introduces the gated recurrent unit(GRU)to make full use of its good time sequence information extraction ability.The experimental results show that CNNGRU MOOCs dropout rates prediction model can effectively predict MOOCs dropout rates,and it is generally superior to other traditional MOOCs dropout rates prediction methods.Thirdly,in view of the problem that traditional MOOCs dropout rates prediction methods only predict dropout rates and ignore learners' information and course information with a very high correlation with dropout rates through learners' clickstream data,the attention mechanism is introduced to combine learner information and course information with the prediction model,so as to improve the generalization ability of the prediction model and construct ATT-CNNGRU MOOCs dropout rates prediction model based on attention mechanism.Finally,ELU function is used as the activation function of GRU network structure in the prediction model to construct ATT-CNNGRU-ELU MOOCs dropout rates prediction model.ELU function can effectively solve the problem of gradient disappearance and has soft saturation,so as to further enhance the robustness and prediction performance of the model.The experimental results show that,compared with ATT-CNNGRU model and CNNGRU model,for ATT-CNNGRU-ELU proposed in this paper,F1-score and AUC value are significantly improved,and its overall prediction performance is optimized compared with the traditional MOOCs dropout rates prediction models.