Research On Academic Early-warning System Based On Improved LSTM Algorithm

It is the fundamental task of higher education to cultivate people by virtue.Realizing the precision of ideological and political education through advanced scientific and technological means has become a research hotspot in today’s rapid development of science and technology.Accurate assistance for students with academic difficulties(students with academic difficulties)is a direction of precise ideological and political education.Existing support strategies in colleges and universities are mostly based on manual counting of unqualified subjects,issuing written warning notices,or using simple correlation algorithms to predict grades.Based on the existing early warning system,this paper conducts research on the improvement of academic early warning system using LSTM neural network.The specific work is as follows:(1)Aiming at the problem of complex types of student behavior data and difficulty in feature extraction,A method for feature extraction of adaptive pre-feedback classification based on multivariate normal distribution is proposed.In order to reduce the impact of missing data on the system’s prediction results,the radial basis function is used to perform high-dimensional mapping on the organized data matrix,and the features of the highdimensional data matrix are extracted by using the front feedback multi-dimensional normal distribution in the high-dimensional space.The data after the feature matrix extraction is sent to different training networks and classification operations.Experiments have proved that classification training can highlight data features and the training results are better.(2)The experiment puts forward an early warning system construction method based on improved LSTM neural network to solve the problem of long-term dependence of student behavior data.The experiment replaces the excitation function in the LSTM neural network with the average function of the adaptive excitation function to achieve the effect of system optimization and increase the calculation rate.The MLP neural network cascaded LSTM optimization network is used to classify and calculate student data,and the results are merged to improve the output accuracy.Experiments show that the improved system model has faster calculation speed and more accurate output results.(3)The system has developed and used the software and actually put it into use.After the system realizes the academic early warning and prediction function,the desktop software is built around the core algorithm,and is designed to realize the functions of graduation probability prediction,credit score query,class score distribution,class academic prediction,etc.The system has been officially put into use in the School of Computer and Information Security,University of G,and has achieved good responsesIn the actual experiment process,the experiment collected the first-year compulsory course scores,consumption information,book borrowing frequency and other data of more than 20,000 people from the four graduates of G University from 2017 to 2020,and cleaned and sorted them.A standard data set is formed.Experimental verification shows that the model’s prediction accuracy rate is stable at 94.21%,and the highest can reach 98.17%.The optimized system model is 2% higher than the average prediction accuracy of the existing early warning model,especially in terms of negative recall rate,in terms of data dependence and result stability.There is also a substantial improvement.