Abstract
The purpose of this paper is to control and judge the big data of students' learning and living conditions in college education. College students' book-borrowing data are mined deeply from three aspects, the multi-source preprocessing of students' borrowing data from university and college libraries, the quantification of students' book borrowing, and academic performance prediction by learning and book borrowing. The data mining technology analyzes and processes students' primary information, score information, and book-borrowing information. Students' book borrowing is modeled and analyzed using the backpropagation neural network (BPNN) algorithm, and the constructed BPNN book-borrowing model's loss function is optimized based on the partial differential equations. The library access control data and book-borrowing data are used for statistics of the learning behavior frequency. Data such as students' stay duration in the learning area and attendance rate are input into the analysis model for experiments, and the average absolute error, the mean square error, and the determination coefficient evaluate the prediction results. The results show that as students' booking borrowing frequency decreases, their scores decrease, and students who often borrow books have strong learning motivations. In Experiment 4, when R-2 reaches its maximum value, 0.594, the predicted scores by the students' book-borrowing model have a high correlation with students' actual scores, indicating that the BPNN algorithm has the best prediction results. The results show that the indicator of students' book borrowing has significantly improved the model's prediction performance, and the borrowed book number and book-borrowing frequency are significant in the prediction model construction.