Analysis And Early Warning Of College Students’ Academic Performance Based On Daily Behavior

Education big data and performance prediction are popular research directions in the education field,they aim at discovering the data features that can affect college students’academic performance by mining and analyzing their daily behaviors and establishing corresponding prediction and early warning models to provide data support and decision reference for college teaching management.However,there are still problems that need to be solved,such as the relatively single data characteristics of daily behaviors,the low accuracy of performance prediction models,and the inability to evaluate the academic early warning situation in advance,which are of great significance for college administrators to improve teaching quality.To address the above problems,this thesis starts from the specific needs of college students teaching management,and designs a set of academic performance analyses and early warning methods for college students based on daily behavior based on the research of a series of traditional data mining and neural network algorithms,which realizes the accurate prediction and classification of college students’final grades and early warning level changes,and provides early warning for academics according to the classification of early warning level,helping college students in the timely adjustment of study status during the semester,reduce the probability of academic warning,and also provide data support for teaching managers.The main work accomplished in this thesis is as follows:（1）Research the extraction and correlation analysis methods of students’daily behavior features,and improve the dimensionality of features related to grades.To address the problem that the daily behavioral data features are relatively single,this thesis investigates how to extract behavioral features from students’usual classroom performance data,one-card consumption data,and library borrowing data,and analyzes the correlation between these behavioral features and students’grades to filter out the features with higher correlation with grades.Through the above methods,this thesis improves the problem of having a single dimension of data features and provides more information and basis for subsequent grade prediction and early warning.（2）Designing a broad learning-based student grade prediction algorithm to achieve accurate prediction of student grades.To address the problem of the low accuracy of the current grade prediction model,a broad learning-based student grade prediction algorithm is proposed.The algorithm model includes three modules:data processing,network training,and grade prediction.The data processing module analyzes students’daily behavior data and extracts effective features,the network training module uses the effective feature data to train and update the network model online,and the grade prediction module realizes fast and accurate prediction of students’grades based on the trained network model.The experimental results show that the broad learning prediction model proposed in this thesis achieves 92.5%and 95.6%in two semesters,respectively,while having the low mean square error,mean absolute error,and root mean square error,which improves the accuracy of student performance prediction.（3）Construct an academic early warning model based on daily behavior to achieve early warning of students’academic status.To address the problem that the academic early warning situation cannot be evaluated in advance,an academic early warning model based on daily behaviors is constructed based on the classification of academic early warning levels using integrated learning algorithms to predict the changes of early warning levels using some of the daily behavioral data of students,to achieve early warning of students’final grades by classification,and to provide timely guidance to students at risk of early warning.According to the evaluation indexes of accuracy,precision,recall,and F₁-Score,the results show that the academic early warning model proposed in this thesis improves the accuracy of grade early warning.