Research on the Prediction of Academic Performance Based on ANN-BiLSTM Hybrid Neural Network Model

With the popularity of online courses and e-learning, a large amount of data on online learning behaviour has been accumulated. How to use these data to predict early students' performance so as to improve teaching effectiveness and learning experience has become an important issue in educational research. Based on this, this paper constructs a hybrid ANN-BiLSTM neural network model by taking advantage of the feature extraction of Artificial Neural Network (ANN) and the advantages of Bi-directional Long Short-Term Memory (BiLSTM) in temporal feature processing. In order to verify the effectiveness of this model, this paper compares it with the Recurrent Neural Network (RNN) model, the Gate Recurrent Unit (GRU), and the ANN-LSTM hybrid neural network (ANN-LSTM) model. Experiments on the UK Open University Learning Analytics dataset showed that the ANN-BiLSTM model had significantly better prediction accuracy and the best predictive performance.