Research On MOOC Dropout Prediction Based On Neural Network

In recent years,massive open online courses(MOOC)have attracted more and more educators and learners because of its flexibility and convenience in the field of higher education.However,it is precisely because of this flexibility and convenience that learners are more likely to drop out.The problem of high dropout rate has become a huge challenge for MOOC platform.In the existing research on dropout prediction,the dropout rate of MOOC is higher than 90%.Therefore,it is necessary to put forward a more reliable and efficient dropout prediction model for MOOC platform to realize the early intervention of educational activities and the development of MOOC platform.In the research of MOOC dropout problem,the traditional machine learning methods were mostly used in the early years.Although they achieved good results,these methods only rely on the shallow feature information extracted by artificial feature engineering,ignoring the relationship between features.Moreover,there are a lot of noise data in the features extracted by the artificial feature engineering method,which affect the prediction effect of the model.In recent years,researchers use neural network to extract features automatically.Although it solves the problem of independence between behavior features to a certain extent,it does not take into account the influence of different learning behavior patterns on dropout and the temporal relationship between learning behaviors,so it is difficult to guarantee the effectiveness of the prediction results.And the data set of kddcup2015,which is commonly used in dropout prediction,has the problem of unbalanced distribution of positive and negative samples,which also has an impact on the prediction results to a certain extent.In order to improve the shortcomings of existing research,this paper uses neural network and improved smote algorithm to improve the performance of dropout prediction.The main content of this paper is as follows:1.Aiming at the problem that the existing artificial feature engineering can’t effectively extract the in-depth feature of learners’ learning behavior and then can’t fully mine the relationship between learning behavior features,which result in the unsatisfactory prediction.In this paper,we proposes a new method named CLNN of dropout prediction,which based on Convolutional Neural Network(CNN)and Long Shor-Term Memory(LSTM)neural network.Firstly,the convolutional neural network is used to extract the deep information of learners’ learning behavior characteristics,and then the long short-term memory neural network is used to further mine the temporal relationship information between learners’ learning behaviors,so that to make the prediction results more accurate.Through theoretical analysis and experimental results,it is proved that this method has a significant improvement compared with the existing dropout prediction methods.2.Aiming at the problem that different learning behavior patterns of learners have different effects on dropout prediction,we proposes a new method named SCGNN of dropout prediction,which based on Squeeze-and-Excitation Networks(SEN)and Gate Recurrent Unit(GRU)Neural Network.The model optimizes the time series feature extraction module of CLNN,and the importance of each behavior feature matrix is displayed,which makes the prediction results more accurate.In addition,considering that the imbalance of positive and negative samples in the data set of dropout prediction affects the performance of the model,we use the improved SMOTE algorithm to train the model by oversampling samples with few samples,which makes the prediction results more accurate.And the experimental results show that the SCGNN model in the early stage of dropout prediction can achieve the same prediction effect as the traditional machine learning model in the whole stage,which further shows the effectiveness of our experimental model.