Under the background of the large-scale popularization of “Internet + Education”, personalized education mode based on artificial intelligence and other information technologies has become the trend of teaching reform development. In this paper, we have collected and pre-processed students’ learning behaviors during the teaching of Probability Theory and Mathematical Statistics courses, and established a large biological database based on these behaviors. The features of learning behaviors are extracted using the SHAP method, and the features are fused and processed by the MFCNN module. The Bi-LSTM method is used to warn students of academic risks. Finally, we construct a visualization system to assist in teaching Probability Theory and Mathematical Statistics. The results of practical application show that with the assistance of the system, the average grade of the students’ Probability Theory and Mathematical Statistics course progresses quite obviously, the post-test grade is significantly higher than the pre-test grade, and the student’s ability of critical reflection also improves. In addition, a significant difference (p<0.05) was observed between the students of the experimental class and those of the control class in the learning motivation posttest. This paper provides new ideas for research related to the big database of biology and academic early warning in teaching, which contributes to the further promotion of teaching reform.