An Exploration of the Early Warning System for College Students' Academic Performance Based on BP Neural Network Driven by Multidimensional Data

This study constructs a Back Propagation neural network–based early warning system for college students' academic performance. Multidimensional data, such as about library visits, consumption, network data, and historical grades, were collected. After the data were standardized and fuzzified, they were input into a single-hidden-layer Back Propagation neural network. The Sigmoid function and gradient descent method were used for training. An analysis of the data from students in the class of 2020 showed that the R2 value of the training set of the model reached .98865. Among the 116 samples in the test set, 105 were correctly predicted, with an accuracy rate of 90.52%, and the error rate was within 0.08. The model can output multilevel early warning results, such as “warning,” “serious warning,” “risk of failing the course,” and “no risk,” thus providing an effective tool for educational administrators to intervene in students' learning in a timely manner and improving the efficiency of educational management and students' learning motivation.