| Educational data mining aims to find the inherent laws from existing educational resources.With the support of machine learning,data mining and other technologies,it can solve some potential problems in the process of education and teaching.Student performance prediction is one of the important research issues in the field of educational data mining,which aims to use the information related to students’ behavior to predict students’ learning performance in the future,such as whether students can pass the exam,test scores,performance ranking,etc.,so as to realize the personalized analysis and guidance.Timely and accurate performance prediction is not only conducive to improving the performance of learners,the teaching quality of teachers and the management efficiency of school administrators,but also conducive to improving the educational and teaching environment.Previously,many scholars have achieved good results in the study of predicting student performance by using the various attributes of students.However,there are still some shortcomings in the current work,which are mainly manifested in the following three aspects: The current work is mostly based on the traditional machine learning methods,while deep learning methods are rarely involved,and the prediction accuracy still needs to be further improved;The current work only considers the influence of the selected attributes on students’ performance,while ignoring the influence of unselected ones;The current work assumes that the attributes have the same impact on all students,ignoring individual differences among students.In fact,different attributes have different influence on the same student’s performance,and different students are affected by the same attribute to different degrees.To deal with the above problems,this article aims to improve the accuracy and interpretability of the student performance prediction model,analyze and use these attributes to predict students’ performance more comprehensively and accurately so as to realize personalized analysis and guidance.The main contribution can be included as follows:（1）We propose a data preprocessing method.This method preprocesses the data according to the characteristics of the attributes,including digital encoding conversion of binary data,numerical normalization,grouping of scores,deleting abnormal data and so on.（2）We propose a method for predicting student performance based on the self-attention mechanism（Self-Attention）,which treats attributes differently and assigns different attention weights,improving the accuracy of performance prediction.（3）We propose a two-way attention（TWA）performance prediction method,which not only treats the influence of each attribute on their scores differently,but also fully considers the relationship between historical grades and the final grades.Firstly,we calculate the attention scores of the attributes on the firststage performance and the second-stage performance.Then we consider a variety of feature fusion methods,so that each attribute can be used more comprehensively and accurately.Finally,we make better predictions of student performance based on the integrated features.（4）We conduct extensive experiments on two public education datasets,and compare with four traditional machine learning algorithms,including support vector machines,logistic regression,Gaussian Naive Bayes,and decision trees.Furthermore,a visual analysis of the prediction results is carried out based on the probability distribution of each attribute on the final grade.The results show that the proposed model can predict student performance more accurately and have good interpretability. |
