The Design And Implementation Of A Learning Early Warning Model

In recent years,because of the rapid improvement of digitization,informatization and intelligence in the education field,learning early warning is becoming a hot topic in the intersection of education and computer.Considering the shortcomings of low early warning accuracy and incomplete early warning methods in the current learning early warning field,this thesis researches the method of student ranking prediction,which is rarely involved at present,and then gives a learning early warning model that integrates scores and rankings.By establishing an accurate and comprehensive learning early warning model,educators can timely provide guidance and help for students according to the early warning signals issued by the model.The main research works of this thesis are as follows:(1)This thesis proposes a Dual Student Comparison Ranking Early Warning Model(DSCRM)for predicting student rankings.Instead of predicting students’ specific scores,the model directly compares students in pairs to obtain the rankings of all students.The thesis shows the structure of the model and explains why it is structured in this way;and then it lists the generation methods of the model’s training set and test set;finally,it shows the complete process of using the model for learning early warning.(2)Based on the previous model,the thesis further proposes a Sample Weight Ranking Early Warning Model(SWRM).In the process of pairing students to generate the model’s input samples,a sample weight is obtained by using the percentage rankings of the two students.The closer their rankings are,the smaller the sample weight is.This allows the model to focus on those samples with a larger impact on the ranking prediction results during training,thereby improving the overall ranking prediction effect.The thesis provides the calculation formula of the sample weight in detail,and shows the method of using sample weights for early warning.(3)Based on the connection between the score early warning model and the ranking early warning model,the thesis combines the two together and finally obtain a Score and Ranking Integrating Learning Early Warning Model(SRIM).While predicting the score of a single student,the model also needs to pay attention to the relationship between the rankings of two students.The thesis shows the detailed structure of the model and the expression of its loss function,introduces its training method;then it describes the generation process of its training set and test set;finally,it displays the learning early warning process combining scores and rankings.The learning early warning models proposed in this thesis have been empirically tested in a specific learning situation.The ranking prediction results of the two ranking early warning models are both better than the corresponding score early warning model and other machine learning algorithms,and the sample weights proposed in this thesis also promote the ranking prediction effect.The early warning model integrating scores and rankings can significantly improve the ranking prediction effect at the expense of slightly increasing the score prediction error,thereby showing students’ learning situations more accurately.To sum up,the thesis obtains the expected learning early warning effect through the attention on the student ranking prediction.