A Proposed Hybrid CNN-RNN Architecture for Student Performance Prediction

Education is one of the important factors for the development of a country, thus, an early prediction system to predict the student performance is needed. With such system, education institutes would have a strong capacity to detect slow learners and investigate the major variables impacting their academic performance and preventing students from dropping out of school owing to poor marks or failing the exam. There are several studies that uses traditional machine learning to predict student performance, however, hybrid deep learning in predicting student performance has never been reported. Whereas machine learning requires feature selection, deep learning models may conduct automatic feature selection in the training model. However, it may suffer from the curse of dimensionality as it collects more and more features from student data. This paper proposes a hybrid deep learning model with combination of Convolutional Neural Network (CNN) and Recurrent neural network (RNN), i.e. CNN-RNN, where CNN captures the local dominant features and reduce the curse of dimensionality and RNN obtains the semantic correlation between features. The results of the experiments indicate that the hybrid CNN-RNN prediction model performs better than deep learning model by 3.16%, where the accuracy increased from 73.07% in ANN to 79.23% in hybrid CNN-RNN.