Analysis And Prediction Of Dropout Behavior Based On School Online

In recent years,the rapid development of online education provides people with a new way of learning.However,with the development of online education platform,the number of registered students is increasing,but the dropout rate is still high.In order to curb the phenomenon of dropout,ensure the development of online education platform,and avoid the waste of educational resources,the research on dropout behavior is urgent.A large number of learning behavior data of the platform provides a basis for the research of dropout behavior.We can build a dropout prediction system according to the information of students’ learning behavior,find the trend of dropout in time,realize the early warning of dropout,and help the platform to provide reasonable and effective high-quality course services according to the learning characteristics and dropout situation of different groups,so as to ensure the learning quality of students To reduce the dropout rate and promote the fair sharing of social education resources.Therefore,this paper analyzes and predicts the dropout behavior of students based on the data of online platform of school.First of all,according to the overall data,the number of people participating in online learning before and after the winter vacation is obviously low,and there is a false learning behavior of playing video but not learning.Secondly,according to Thorndike’s law of learning,combined with the characteristics of data,time series features,event features and behavior features are selected and processed.According to the characteristics of the data,the advantages and disadvantages of neural network,logistic regression,naive Bayes and decision tree model are compared,and the decision tree and random forest model which are suitable for structured data and have strong logic are adopted to predict the dropout behavior.Finally,by setting the time series feature as the initial feature,adding the event feature and behavior feature in turn to construct the model three times,mining the important features,adopting the grid search method to optimize the parameters,strengthening the generalization performance of the model,using the ten fold cross validation combined with four evaluation indexes of accuracy,accuracy,recall rate and F1 value to evaluate the prediction results,so as to ensure the accuracy of the prediction results Effectiveness.The results show that after adding behavioral characteristics,the prediction effect is significantly improved,the four evaluation indexes of decision tree model are improved by about 60%,and the accuracy,recall rate and F1 value of random forest model are improved by about 40%.The recall rate of random forest is as high as 0.9785 and F1 value is as high as 0.9710.Moreover,the evaluation indexes of random forest are basically higher than those of decision tree.The research results of this paper show that from the three aspects of behavior characteristics,grid search and random forest model,it can help to realize the accurate prediction of class dropout behavior,provide a new idea for the analysis of class dropout behavior,and can be applied to the actual online education platform to realize the early warning of class dropout and solve the problem of class dropout on online platform.