Research On Learning Effect Prediction And Feedback Mechanism Based On Behavior Analysis

With the steady advancement of education informatization in China,the user scale of online education platforms is increasing,but at the same time,problems such as low learning efficacy of learners,confused learning status,and difficulties in teacherstudent interaction due to the online teaching environment have been exposed.Researchers have found that the interaction between learners and the platform generates a huge amount of behavioral data,which contains the implicit behavioral characteristics of learners.Using these characteristics,constructing models with appropriate data mining algorithms,and exploring the hidden laws behind the behavioral characteristics are the unique advantages of online education compared with traditional education.In this thesis,we dig deeper into online learning behavior data for building learning effect prediction models,establishing learning feedback mechanisms,and providing timely feedback and real-time interventions in the learning process,which will help improve the disadvantages of online education environment such as insufficient learner selfdrive and inconvenient information interaction,and enhance the level and quality of education informatization.Based on student behavior analysis,this thesis is carried out in three aspects:learning behavior feature screening,student group clustering,and learning effect prediction,and uses the research results to design a learning feedback mechanism.The research work in this thesis mainly includes the following four aspects: 1.Deep mining online learners’ behavioral data,extracting behavioral features related to learning effects from them to make an experimental dataset,and using the Gini coefficient index of the Random Forest(RF)important feature selection algorithm to filter out the higher importance learning behavior categories among the learning behavior feature terms.2.Improving SOM self-organizing neural network by using the systematic clustering method,and obtaining different behavioral clustering analysis of the features to obtain the behavioral characteristics of students classified in different achievement intervals.3.The LSTM-based learning effect prediction model is proposed to obtain the temporal characteristics of behavioral data with the help of long short-term memory network(LSTM)to reach the purpose of stage prediction of learning effect in weeks.For the problem of data imbalance in the dataset,the KM-SMOTE oversampling method is used to improve the prediction effect of the prediction model effectively.4.Based on the above experimental results,four types of learning feedback mechanisms are proposed: learning behavior association analysis、 learning recommendation system、learning warning system and learner portrait design.The conclusions and innovations of this thesis are shown below: 1.The educational dataset features are filtered by the feature importance score of the random forest algorithm,which improves the computational efficiency and reduces the interference of irrelevant terms to enhance the subsequent model prediction.2.The clustering quality of the SOM neural network clustering algorithm,which has been improved by the systematic clustering method,is improved compared with the effect of the traditional clustering algorithm,and the experiments divide the student groups into five cluster classes according to the learning behavior differences.The experiments divide the student groups into five cluster classes based on the differences in learning behaviors,and propose different teaching strategies for different groups of students.3.With the help of the long short-term memory network(LSTM)to build a learning effect prediction model,the implementation results show that the model proposed in this paper exceeds the traditional machine learning methods,and the learning prediction effect is significantly improved.4.Using the KM-SMOTE oversampling method improved based on SMOTE and K-means algorithms,it can effectively prevent interpolation flooding.method,which can effectively prevent interpolation pan-boundaryization and solve the problem of unbalanced data categories of learning behavior samples,is more helpful to improve the learning prediction effect of unbalanced data set models.