Research on Detection Technology of Abnormal Data in College Physical Education Network Teaching Test Results

Abstract

In recent years, with the rapid development of big data technology, more and more data are continuously generated with the summary of university systems. Therefore, how to use these educational data to provide more scientific decision-making information for university information builders is very important. This research collected various educational data through the network teaching system and campus information system. After that, the collected data was used to analyze the students' online behavior and to excavate valuable behavioral characteristics. In addition, the experiment also proposed indicators to describe students' behavior, such as network course behavior, network viscosity and life regularity, to provide the basis for the subsequent abnormal performance prediction model. Finally, the experiment used DBSCAN algorithm based on distance optimization for clustering analysis, and constructed a NA model based on multiple classifiers. The research results showed that when, the SC value was 0.711, which was the optimal solution of D-DBSCAN algorithm. At this time, the corresponding number of clusters was 4. When N=2, that was, the base classifier of NA model is composed of C4.5 model and SVM model, the prediction accuracy and time consumption are the most appropriate. The accuracy, recall and F1 values of NA model were 98.16\%, 97.26\% and 0.958 respectively, which was better than that of single model. To sum up, the NA model based on classifiers proposed in the study had higher accuracy and better model performance, can effectively reflect students' academic level, and could provide accurate abnormal performance data for college sports online teaching tests.