Student Achievement Analysis And Prediction Based On Campus Data

Student achievement prediction is an important research direction of educational data mining.With the enrollment expansion of colleges and universities,the number of students keeps growing,which leads to the shortage of teaching resources in colleges and universities,and the analysis and prediction of student achievement has attracted more and more attention.Student achievement prediction aims to predict students’ future academic performance through all kinds of data information related to students,such as campus data information.The construction of the achievement prediction model can provide effective help for teachers to differentiate teaching and guide students’ course learning,and avoid some students from repeating grades,failing grades and even dropping out.Before that,many scholars have done a lot of research on the task of predicting student achievement and have made some achievements.However,there are still some problems and shortcomings in the existing research on student achievement,mainly in the following three aspects: Most of the current research is based on the traditional machine learning algorithm and only uses a single machine learning algorithm for grade prediction model,which has low prediction accuracy and weak generalization ability.Current researches rarely involve deep learning related algorithms,which can not fully utilize and learn higher-order features of various attribute information,resulting in poor prediction effect.Current studies all assume that each influencing factor in campus data has the same influence on students’ academic performance,but in fact,different influencing factors have different influence degrees.Moreover,current studies also ignore the important correlation between historical scores and final scores.In view of the above problems,this paper focuses on traditional campus classroom education and uses campus data to predict students’ final grades.The grade prediction model constructed mainly improves the prediction accuracy and interpretability of the model.The main research contents of this paper are as follows:(1)Student achievement analysis and data preprocessing.First of all,the final grades and attribute characteristics are drawn scatter plot to analyze the correlation between the final grades and each attribute.Conduct data preprocessing for each attribute feature of campus data of students,including grouping students’ grades to classify grades,carrying out numerical normalization for each attribute characteristic value,deleting abnormal sample data,and carrying out coding conversion for attribute characteristic value.(2)Construct the student final grade prediction model based on Stacking fusion.Firstly,random forest algorithm,GBDT algorithm and XGBoost algorithm were used to establish a single model to predict the performance.In order to further enhance the prediction effect of the model,this paper integrates multiple traditional machine learning algorithms based on the Stacking fusion mechanism.A two-layer Stacking framework is used.The base learner uses random forest,GBDT,and XGBoost,and the secondary learner uses logistic regression algorithm.Compared with the achievement prediction effects of three single models,each classification model index of the Stacking fusion model increases by more than 10%.(3)Construct the SAAA-Attention final grade prediction model based on the multi-channel attention mechanism.There are three attention mechanisms in this prediction model,one of which is Self-Attention mechanism,which assigns different attention weights to the characteristic values of each attribute,reflecting the different degrees of influence of each attribute characteristics on the final grades.The other two are the Attention mechanism for historical performance,which calculates the attention weight of non-academic attribute features on historical performance in the first stage and the attention weight of each non-academic attribute feature on historical performance in the second stage respectively through the attention mechanism,and effectively makes use of the correlation between historical performance and final performance.Finally,the three feature vectors obtained by the three-way attention mechanism are fused and input into the BP neural network to predict the categories of students’ final grades.Compared with the prediction effects of three single models and the Stacking fusion model,the SAAA-Attention model increased each classification model index by more than 8%.(4)According to the constructed SAAA-Attention prediction model,a student’s final grade prediction system was developed and realized by using Java and Python languages,front and back end development framework and deep learning framework.The system can use campus data information to accurately predict students’ final grades,and can use word cloud image to visualize students’ campus data information.