Research On Learning Effectiveness Prediction Based On Online Learner Data

With the popularity of online learning platforms and the wide range of learners,relevant research at home and abroad is becoming more and more abundant.However,the prediction is mostly limited to the research of behavior data;The prediction analysis has the shortcomings of single index and homogenization;The selection of prediction algorithm is lack of pertinence,lack of consideration of obvious timing and high imbalance ratio data,and the formation model is too complex.In view of the above shortcomings,this paper makes the following research based on the data of EDX and school online platform from the dimensions of research objects,the selection of indicators of learning effectiveness and the direction of algorithm selection:(1)In terms of the selection of research objects and analysis indicators of learning effectiveness,this paper takes learner data as the research object,comprehensively considers the general attributes and online behavior attributes,defines learning effectiveness from multiple angles,and enriches the selection of prediction indicators.Based on the ed X platform data,through attribute reduction and Research on the relationship between attributes,this paper finds that the certified in the ed X platform data learning effect is predicted by the learner general attribute(Lo E?DI)and multiple online behavior attributes,and good results are obtained;Due to the rich data attributes of ed X platform,its learning effectiveness is divided into three evaluation indicators: whether to obtain a certificate,grade and total?time;According to the online platform data of the school,the learning effect is defined as whether there is a withdrawal behavior in the last week.(2)Based on the characteristics of prediction indicators of learning effectiveness,this paper selects appropriate prediction algorithms for them.Experiments show that the certified in the data learning effectiveness index based on ed X platform has achieved good results by using support vector machine algorithm for binary classification prediction,while the other two indexes belong to multi classification problems,which is more suitable to use the fusion of machine learning algorithm for prediction.(3)In order to take into account the locality and timing characteristics of data,this paper proposes two serial algorithms: WCADT algorithm based on Convolutional Self Encoder-Long Short-Term Memory-Decision Tree and WCLSRT algorithm based on Convolutional Neural Network-Long Short-Term Memory-Random Forest algorithm.Convolutional Self Encoder and Convolutional Neural Network are used to automatically extract the local features of online learners' behavior,and retain the temporal features.Then,the temporal features are combined by Long Short-Term Memory,and finally predicted by machine learning algorithm.Based on the online platform data of the school with obvious time series characteristics and high imbalance ratio,AUC is selected as the evaluation index,which has been improved on the basis of the existing model.The research shows that through the analysis of data dimension and learning effect,the selection of indicators and the improvement of algorithm selection direction,we can more fully study the value behind the data,timely adjust learners' behavior through prediction,improve learning efficiency and promote the better and faster development of online learning mode.