Prediction of Students' Learning Behavior Under the Blended Teaching Mode Based on Big Data Analysis

The blended teaching mode combines the advantages of traditional face-to-face teaching and modern online learning, and provides students with personalized and diversified learning experiences. By collecting and analyzing students' data in the process of blended learning, including learning duration, participation, homework completion and online test scores, this study constructs a prediction model to predict students' learning performance, effectiveness and potential difficulties. The study employed various machine learning algorithms, including Decision Tree (DT), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and Long Short-Term Memory Network (LSTM), to conduct an empirical analysis of the learning behavior data of 1,000 college students. The results indicate that learning duration and visit frequency are key features for predicting learning outcomes, showing a significant correlation with academic performance. Model performance evaluation revealed that the LSTM model performed best in terms of accuracy, recall, and F1 score, achieving an accuracy of 0.85. These findings provide a scientific basis for personalized teaching, helping educators optimize their teaching strategies, improve educational quality, and offer customized learning resources and activities to students.