Font Size: a A A

Research On Mining And Analysis Of College Students' Behavior And Performance Prediction

Posted on:2022-08-28Degree:MasterType:Thesis
Country:ChinaCandidate:T Q LeiFull Text:PDF
GTID:2517306515456334Subject:Master of Engineering
Abstract/Summary:PDF Full Text Request
Educational informatization continues to develop,which has attracted more attention.Building campuses with information technology shows that the era of smart education has come.There are a lot of educational resources,and the value of mining data is too little.Aiming at the above,this studies behavioral data analysis,analysis of the conjunction between behavioral data and performance data,and student performance prediction to mine useful information,strengthen learning effects and management levels.The main tasks are as follows:(1)Multi-factor improved K-means clustering algorithm.In view of the traditional Kmeans clustering algorithm's complex calculations in the iterative process,clusters' number has to be artificially determined,sensitivity to the initial clustering center,susceptibility to outliers,an optimized K-means combining multiple points factors is proposed based on the optimized isolated point detection methods and maximum-minimum distance idea as well as heuristic methods.Results demonstrate that the optimized algorithm is very stable,with an average improvement of 9.17% in accuracy,an average reduction of 4.16 in he number of iterations,an average reduction of 5.70% in the DBI index,and an average increase of 5.22%in the SC index compared with the FECA clustering algorithm.(2)Analysis of student behavior based on clustering.In view of the involuted and diverse life and study behaviors of university students,which is not convenient for behavior analysis,student behavior description indexes are established,and student behavior data in the description indexes are clustered using an optimized algorithm.Different student behavior categories on each index are obtained to explain the behavior feature of each category of students,and summarize the behavior feature labels of different categories of students.(3)Association rule mining based on Apriori.In response to the problem that the operation efficiency of the pruning step of Apriori algorithm is low and the limitations of the traditional association rule metrics.The Apriori pruning operation is optimized by using the directional scanning method,and the traditional association rule metrics are optimized by introducing the effectiveness metrics.Based on the clustering results of student behavior data,the optimized Apriori is used to dig the rules among students' course achievement,doings and grades,and analyze the correlation among them.(4)Student performance prediction based on Attention-Bi LSTM.Aiming at the problem that traditional performance prediction methods ignore the different degrees of influence of different behavioral features on student performance,and considering that behavioral data in different periods have different degrees of influence on student performance,thus the achievement prediction problem is converted to a classification problem sequentially.The Attention-Bi LSTM achievement prediction model is created by integrating the attention mechanism and the long and short-term memory neural network.Experiments demonstrate that the prediction model raised in this paper has improved accuracy by 15.72% and 7.21%,respectively,compared with the Logistic Regression model,which has better prediction effect in the benchmark model,and the long and short-term memory neural network model without the attention mechanism,which effectively ameliorates the prediction quality.
Keywords/Search Tags:student behavior, cluster analysis, association rule mining, performance prediction, long and short-term memory network
PDF Full Text Request
Related items