Network Behavior Analysis With Distanceoptimized DBSCAN And Prediction Of Academic Performance Using N-Adaboost Model

As more and more education data are generated by the aggregation of university application systems and hardware equipment,how to dig out more scientific and accurate information for decision makers at all levels from these big data has brought a major challenge to university information builders.Academic performance analysis is an integral part of educational data mining and provides a basis for schools to evaluate students in a comprehensive and objective manner.Accordingly,college administrators can grasp the characteristics and patterns of student behavior,and promote individualized education,abnormal behaviour can be detected and intervened timely to improve the quality of teaching management and student services.At present,domestic and international research on the analysis of academic level in higher education is based on simple data and simple models with a shallow level,and single field oriented.Some data focus on the consumption data of campus card,and some data on the E-education platform,but there is no study on the impact of online behavior and campus card data on academic performance.In view of defects and deficiencies of the existing research fields,this paper focuses on online behavior for the first time,takes the undergraduate students of a university as an example,and analyzes the data of students'online behavior,campus card data and academic performance.Empirical analysis shows that a combination of online behavioral and campus card data can reflect academic levels more effectively and produce more accurate predictions.The specific work of this paper is as follows:?1?Construct a"portrait"feature library of student behavior,and propose"Three Aspects"of network behavior,network viscosity,and life regularity to effectively portray student behavior;For the traditional DBSCAN algorithm,a distance-optimized DBSCAN algorithm is proposed,which optimizes the field radius?,selects the initial radius?by calculating the average distance between the minPts sample points,and dynamically adjusts the field radius using the distance coefficient between the core points and the sample points which in the?field during the clustering process to speed up the clustering convergence;This clustering algorithm is performed on the student behavior indicators from"Three Aspects",and the groups of students with different behavior characteristics are obtained;Empirical analysis shows the distance-optimized DBSCAN algorithm has 9.2%better clustering effect and accuracy in student behavior,and its comprehensive performance is better than the traditional DBSCAN algorithm.?2?Visual analysis of factors affecting academic level.The performance patterns and variability of the factors affecting academic performance among different groups of students are analyzed and described in detail,and the ANOVA F-test is used to screen out the characteristics that have a significant impact on academic performance.?3?An heterogeneous N-Adaboost algorithm model based on multiple classifiers is proposed.The base classifier consists of N classifiers H₄)?x?which results are generated by H₄)?x?voting,and the final prediction results are obtained by iteratively updating the sample weights on???.The accuracy of the N-Adaboost model based on multiple classifiers in predicting academic performance of different groups of students was shown to be significantly improved,reaching 73.29%?"pass&fail"group?,73.74%?"excellent&non-excellent"group?and 81.36%?"excellent&fail"group?,all higher than other classifiers.The effect of different N on prediction performance and accuracy was analyzed,and it was found that the model consumed much more time than the prediction accuracy at N>3.The model improves the types of base classifiers of the traditional Adaboost model,complementing the advantages of each classification model,improving the"short board"of the algorithm and achieveing effective improvements in performance and prediction accuracy.?4?Design and implementation of an early warning system for schooling.Based on the N-Adaboost algorithm model and Django MVC framework,the early warning system is designed with a three-tier architecture:data access layer,data processing layer,data display layer.The system contains four functional modules:basic information module,network behavior module,visualization module and early-warning module,and the functional pages and workflow of the system are explained using an undergraduate student in a college of the university as an example.The system realized in this paper provides school administrators with various functions such as students'personal information,network behavior and academic performance early-warning,and has been applied in the Student Affair Department,which has good practical significance.