The Prediction Of University Students’ Academic Achievements Based On Attention Neural Network

In recent years,with the continuous development of ICT technology,the construction of digital campus for many years,the rich campus information system is closely combined with the study and life of college students.College students also produce a lot of data in such campus study and life and store it,but the data deposited in the system as a valuable asset has not yet fully released its value.Nowadays,the urgent problem for colleges and universities is how to make good use of educational data resources with the help of science and technology to provide scientific guidance and valuable decision-making support for teaching,learning and working in colleges and universities.Among many research directions,the urgency of academic performance prediction is particularly prominent.This is because the academic performance of college students directly reflects the students’ learning achievements and affects their future graduation,promotion and employment.In addition,the level of students’ academic performance is also an intuitive evaluation standard for the teaching quality of colleges and universities.College management departments can predict their future academic performance through academic performance prediction methods.If it is predicted that students will have the risk of failing a grade,repeating a grade or even dropping out of school,and the reasons are analyzed through data mining,student managers can carry out appropriate academic early warning work according to the analysis results,and finally achieve the goal of ensuring talent training.Therefore,it is of great significance to study an accurate method for predicting students’ academic performance and find out the influencing factors of their studies for improving teaching quality and strengthening student management.For these reasons,academic performance prediction has always been a hot spot in the field of education research.The author believes that there are many problems in the existing research on academic performance prediction through work experience and consulting relevant literature:for example,the factors that have been studied that affect academic performance are relatively simple and subjective,there are few research on academic performance prediction for college students,the prediction results are not accurate enough and have no guiding significance for academic early warning,etc.For this reason,the author decided to conduct academic performance prediction research based on the multi-dimensional behavior data generated by college students’ learning and life on campus,and to find a more suitable prediction model for college scenes to optimize and achieve more accurate prediction.There are many factors that affect the academic performance of college students.When analyzing the influencing factors,the more comprehensive and authentic the analysis of the influencing factors,the more we can discover the real reasons that affect the academic performance.For this reason,the data set used in this paper is from the real data of a domestic university from 2016 to 2020.The data covers multi-dimensional behavioral data,which solves the problem of data collection methods that only rely on questionnaires and other highly subjective and ignore unknown factors.Analyze the factors that affect college students’ learning,and select the factors with high relevance.When analyzing the correlation of influencing factors,it was found that students’ scores were significantly related to the frequency of eating breakfast,attendance and learning time.The student work and teaching departments of colleges and universities valued these factors and gave positive guidance to achieve the goal of reducing academic risk.At the same time,in the study,it was found that the length of online time that traditional cognition had a negative impact on learning had a strong positive correlation with the performance.After in-depth exploration,it was found that this was related to the popularity of more and more online courses such as Muke in recent years.In terms of research methods,different from the models used in previous studies,in order to better deal with the long time series problem,the author uses a neural network prediction model with attention mechanism Informer to predict.This method extracts the input sequence information through the self attention mechanism in Informer model,which can model the complex nonlinear relationship between the influencing factors,and improve the prediction accuracy,Moreover,the decoder can shorten the length of the input sequence,thereby saving the memory consumption of the encoder and shortening the prediction time.The author trains,verifies and tests the model through real data,and conducts comparative experiments with long short term memory(LSTM)and recurrent neural network(RNN).The experiments show that Informer model algorithm is effective and has higher prediction accuracy.