Online Performance Prediction Combined Prior Knowledge and Deep Learning Models

With the rapid development of information technology, an increasing number of student learning activities are taking place in online environments. Analyzing online learning behavior data can further improve teaching and assessment methods, enhance learning outcomes, and promote independent learning. However, it is a challenge for teachers to evaluate learning status and academic performance of each student through online learning systems. This study aimed to explore an efficient method for assessing student academic performance. A performance prediction method combined prior knowledge and deep learning models was proposed. Before putting learning behavior data into deep learning models, a weight matrix for learning behaviors was reconstructed. And the learning behavior features derived from this weight matrix along with the raw learning behavior data were imported into a custom Dense layer network. Additionally, an attention mechanism was incorporated into the model to improve model prediction accuracy. Experimental results showed that this method could effectively predict academic performance, identify high-risk students who may need assistance and improve the quality of online teaching.