Analysis Of Learning Behavior And Prediction Of Learning Achievement Based On Campus Big Data

The study of student behavior in the campus scene can effectively understand the characteristics of students' learning and living behaviors,and facilitate the teaching in accordance with their aptitude in the teaching process.It is currently a research hotspot in this field.The vigorous promotion of digital campuses has prompted researchers to abandon the traditional data collection methods represented by questionnaires and start to look at digital campus data,broadening the dimensions and universality of related research.At the same time,the development of technologies such as big data and artificial intelligence has completely overturned the technical means(represented by statistical analysis)commonly used in related research fields,providing new directions and ideas for the development of this field.The analysis of student behavior based on campus big data helps to understand students' behaviors and growth patterns,and on this basis,adapts them to personalized learning feedback and other services.It has important theoretical research and practical application value.Although this field has gradually attracted the attention of researchers at home and abroad,in general,the relative research results are still relatively small.This study takes the prediction of academic performance as the starting point and conducts an in-depth analysis of student behavior based on campus big data.The study consists of three parts:Firstly,this study aggregates multi-dimensional,large-scale(N=683)student behavior data.In order to perceive the behavior of students-on a large scale and comprehensively,this study integrates multi-source data in a digital campus environment,covering educational administration systems,campus card,campus WI-FI,etc.On the basis of privacy protection,a unified data format and specifications were formulated to aggregate and fuse multi-source data,which resolved the mutual isolation between different sources and different types of data on the same learning subject on the campus.Secondly,this study comprehensively uses techniques such as time-frequency analysis and non-linear analysis to systematically quantify student behavior data from the perspective of learning diligence and behavior regularity.On the one hand,based on traditional statistical analysis and other methods,and from the perspective of non-linear analysis,two metrics of approximate entropy and change-complexity are creatively proposed to further quantify the regularity(e.g.complexity)of student behavior.On the other hand,in order to more comprehensively evaluate the performance of the newly proposed metrics,this paper proposes the FSA(Feature Scores Average)metric from the perspective of "weighted" analysis.The experimental results show that the newly proposed metrics are greatly improved compared with the traditional method.And then,this study further explores the mechanism of influence between student behavior and academic performance.With the help of visual analysis methods(e.g.histograms,color histograms,etc.)and Pearson correlation coefficients,a systematic analysis and evaluation of relevant behavioral characteristics is obtained,and We get the behavioral patterns of students in different grades.Finally,based on machine learning algorithms,this study builds a regression model for learning achievement prediction.In this paper,we perform ablation experiments to explore the performance of the model and the necessity of different features in model prediction.On the one hand,after comparing the differences and model effects of four different algorithms of SVR,RF,GBDT,and XGBOOST,this study chose the XGBOOST algorithm with the best effect to build a prediction model.The experimental results show that the MSE between the predicted result and the actual score is 0.0092,and the R2 is 0.4742.On the other hand,the necessity and contribution of different characteristic metrics in the model are analyzed.The experimental results show that the behavior characteristics quantified in this paper have effectively improved the predictive performance of the model,that is,the behavioral characteristics are very necessary in the construction of the prediction model.