Optimization of blended teaching mode for college courses based on deep learning and BOPPPS model

Abstract Background Blended Teaching (BT), integrating online digital learning with traditional classroom instruction, is vital in modern higher education. However, traditional BT models often lack personalization, resulting in reduced engagement and uneven learning outcomes. Objective This research advances predictive modeling and analytical assessment of BT in college courses by incorporating the Bridge-in, Objective, Pre-assessment, Participatory Learning, Post-assessment, and Summary (BOPPPS) framework with deep learning methods to assist data-driven examination of student learning patterns. Methods Data collection records engagement indicators, quiz results, and participation levels throughout the English course. Participatory learning incorporates group projects and virtual labs to encourage active learning. The collected data is analyzed using the Efficient Elman Memory Neural Network (E2MNN), which merges the temporal memory capabilities of the Efficient Elman Neural Network (E2NN) with the sequence-learning strengths which improves convergence speed and reduces computational complexity in modelling learning dynamics of the Long Short-Term Memory (LSTM) network enhances the network’s ability to learn complex, time-dependent patterns in student performance data. Results E2MNN can be used to identify learning patterns and predict weak points well, facilitating the analytical assessment of student performance. Post-assessment and feedback are delivered through AI-powered personalized evaluations. Python software was employed to develop the E2MNN model, achieving 96.89% accuracy, 94% precision, 92% recall, 90% F1-score, 2.94 RMSE, 0.03 MAPE, 16.45 MSE, and 0.025s training time, demonstrating excellent predictive performance. Conclusion Using SPSS version 28, paired t-tests showed significant pre–post differences in performance (p