Improving Student Learning Risk Detection: A Multi-Level Early Warning Framework with Adaptive Mechanisms

Early warning of learning is important for reminding students to improve themselves and helping teachers to optimize their teaching plans. Researches on early warning of learning have not effectively modeled the fluctuations in students' learning states and have not provided multi-level warnings for students at different levels. To address these issues, a new early warning method is proposed, combining cognitive diagnosis with learning behavior analysis to predict risks in the learning envi-ronment. We use convolutional neural networks and long short-term memory networks to explore students' potential learning features, and attention mechanisms are used to enhance feature extraction. We use Adaboost algorithm to predict students' learning performance. To prevent overfitting and improve data accuracy, an adaptive mechanism is added to the output layer. Based on data analysis results, multi-level early warning of learning are provided for students. Finally, experiments show that the proposed method can effectively and accurately predict at-risk students.