An Analysis Of Campus Student Behavior In Network Perception

The development of Internet of things,cloud computing,big data analysis and machine learning technology has affected all aspects of social life and promoted the development of smart campus.Smart campus aims to create a smart campus environment,so as to make the management of students,teachers and school administrators more scientific and convenient.Smart campus environment is conducive to the development of the school,but also has an important impact on the personal development of students.Smart campus should have the ability to dig and analyze students'behavior.The daily behaviors of students can reflect the status in the learning process,the quality of teaching and whether the school's management mode is good.The research on them can find the problems in the education work and put forward the corresponding solutions.In addition,understanding students'behaviors can help to find hot spots on campus,improve campus security and optimize campus services.With the popularity of mobile smart devices,the network can perceive a large amount of data and information.Based on these data,user behavior can be obtained and rules can be found.In this paper,the behavior data of students in Bupt campus are acquired through the developed WiCloud system,and the research is carried out from the following three aspects:clustering of students'behavior patterns,analysis of learning atmosphere and prediction of academic performance.1.Cluster analysis is carried out on students' behaviors,and students'behaviors are subdivided into different types to obtain the characteristics of different types of student groups.Through the two processes of spatialfeature processing and temporal feature processing,the characteristics of students' behavior patterns are extracted.By taking advantage of the strong interpretability of K-means clustering algorithm,the characteristics of behavior patterns are clustered and analyzed.The results show that students' behaviors can be classified into three types:enclosed type,active type and study-oriented.The peak of time distribution for enclosed students is in dormitories,where they spend an average of 58%of their time.The peak time distribution for study-oriented students is in the laboratory,where they spend an average of 42%of their time.The time distribution of active students has no obvious peak location,and their stay time is relatively balanced.Enclosed students accounted for 45%,active students accounted for 31%,learning students accounted for 24%.We can also conclude that boys are more active type than girls.The proportion of enclosed students in undergraduate students is less than that of graduate students.2.Analyze the changing rules of study atmosphere.Using the proportion of students' stay in the laboratory as the measurement of learning atmosphere,the numerical analysis shows that from 2015 to 2017 the proportion of students' stay in the laboratory increased from 22.0%to 25.8%.The study atmosphere at the end of the semester is better than that at the beginning of the semester.The students have better study atmosphere in the autumn term than in the spring term.The study atmosphere of male and female students shows the same change trend,and the floating size is basically the same.Graduate student's study atmosphere is quite stable,while undergraduate student's study atmosphere fluctuates greatly.3.Student achievement prediction.The prediction of academic performance is abstracted as a classification problem.In the process of constructing student feature data set,feature extraction is carried out from two aspects:the degree of effort and the regularity of life.Common classification algorithm models were established,including Support Vector Machine(SVM),decision tree,logistic regression,K-nearest neighbor and neural network model combining Attention mechanism for comparison and verification.Further experiments were carried out using a combination of five predictors.The results show that:(1)There is a correlation between students' daily behavior and their academic performance.Among the common classification models,SVM model has a good applicability in the scenarios where student behavior predicts academic performance,with an accuracy rate of 72.9%.The model based on neural network can further improve the accuracy,which is 75.5%.(2)Using the proportion of students' staying time for prediction also has the accuracy beyond the baseline,but the accuracy is relatively low.(3)There is a positive correlation between the proportion of time spent on weekdays in learning places and academic performance.