MOOC Learning Behavior Mining And Dropout Prediction Method Research

With the development of modern education and the rise of the Internet,MOOC(Massive Open Online Course)online learning has become a essential learning means.Although there are considerable number of learning participants and many advantages such as being free,convenient,and efficient on the MOOC learning platform,only a few learners can complete the final course,the low complete rate has seriously hampered the development of MOOC.The large amount of learning behavior data should be fully used to solve the problem of low MOOC complete rates.In this study,learners who did not study for a week are categorized as dropouts,according to the behavior data of learners in MOOC,the study of this article consists of two major components.The first part is mining the learning behavior of MOOC,which discovers and analyzes the learning behaviors of different learners through clustering and association rule mining.The second part is the dropout forecasting section,which uses the proposed predictive model to train through the learner's learning behavior data to predict the learner's dropout behavior.It mainly completed the following work:(1)A research on clustering based on learning behavior data.Through clustering,learners were divided according to the active degree of learning behavior in the MOOC learning process.Firstly,aiming at the problem that the K-means algorithm is sensitive to the initial center point of the cluster and has high computational complexity,a clustering algorithm based on the idea of distance sorting to improve the initial point selection is proposed.Using the four standard datasets of the UCI to conduct comparative experiments,the results show that,Compared with the K-means algorithm,the accuracy of the improved algorithm was increased by 18.51%,6.82%,30.77%,and 15.79%,respectively,and the time reduction was 20.45%,14.67%,30.46%,and 25.07%,respectively.Secondly,11 dimensions of learning behaviors representing learners are reduced by PCA,and learners are clustered based on learning behavior data and using the proposed improved algorithm.the contour coefficient method was used to determine the number of clusters.The experimental results showed that when the number of clusters was 3,the profile coefficient reached the peak,so the best clustering effect was to gather learning into three categories.(2)A research on Mining Association Rules Based on Learning Behavior Data and Clustering Results.In the research,fuzzy association rule mining algorithm is used to mine the association rules between different learners' learning behaviors from two perspectives.The first perspective is to mine fuzzy associations between the learning behaviors of the two types of learners with the most difference in positive degrees among the above clustering results.The second perspective is to mine the corresponding rules of fuzzy association between learners' learning behavior at different times.Analyze the fuzzy association rules excavated from the above two angles.(3)A study of prediction of dropouts based on learning behavior data.Due to the continuous acquisition of characteristics for each student for a period of time,so dropout prediction was essentially a time series prediction problem.The dropout prediction was used as a sequence classification problem to propose a time model to solve it,namely a recurrent neural network(RNN)model with long-term short-term memory(LSTM)cells.The dropout prediction model was verified according to the definition of dropouts given.Comparing trained model with support vector machine(SVM)and logistic regression(LR),it was found that the LSTM model was better than other models in predicting the learner's dropout.