A Performance Prediction Model Based on Combined Autoencoder

Since most performance prediction models fail to make effective use of the essential characteristics of student performance data, this paper builds a combined autoencoder student performance prediction model, named HS-BP, which combines marginalized denoising autoencoder and stack sparse autoencoder to perform unsupervised feature learning on the student's historical performance data and behavior data, and connects the BP neural network on the top layer to achieve student performance prediction. Experimental results show that the proposed HS-BP model has higher prediction accuracy than other shallow models without feature learning.