Construction of College Students' Network Behavior User Portrait Based on Stacking Machine Learning Model

With the rapid development of Internet technology, college students' network behaviors are becoming more and more diversified, which brings new challenges and opportunities to educational management, psychological counseling and campus security. In this study, Naive Bayes, Random Forest, Support Vector Machine and Logistic Regression algorithms from machine learning methods were selected as the first-layer learners. Decision Trees algorithm is used in the second layer learner and an algorithm model based on Stacking machine learning is constructed. Experiments show that compared with simple model machine learning algorithms such as Naive Bayes and Random Forest, the classification accuracy is increased by about 3%, and the recall rate and F1 value are both increased by 4%-5%. Through the research on the correlation between college students' Internet behavior and academic performance, it is found that there are significant differences in college students' Internet behavior preferences in different dimensions. The main online behaviors of excellent students are learning (27.95%), video (18.51%), social networking (13.44%), search (10.91%), and reading (10.57%). Underachievers major Internet behavior for (32.19%), video game (17.36%), social (13.78%), search (8.70%), music (6.40%). This study not only enriched the theoretical system of network behavior portrait, but also provided powerful data support for university education management, personalized recommendation and academic early warning.