Effective Deep Learning Model to Predict Student Grade Point Averages

Abstract

The main objective of this paper is to provide an overview of the deep learning techniques that can be used to predict a student's performance as compared to other traditionally used machine learning techniques. In our research, we developed feed forward neural networks and recurrent neural networks for developing a model to effectively predict the student GPA. The recurrent neural networks gave greater accuracy as compared to feed forward neural networks, as they have memory and take into consideration the consistency of the student performance. The main contribution of the paper is that we have compared various recurrent neural architectures such as single hidden layer long short-term memory network, long short-term memory network with multiple hidden layers, and Bi-directional long short-term memory network with multiple hidden layers. We compared these techniques with root mean square error as the parameter of comparison and found Bi-directional long short-term memory network to have the least error of 8.2%. A comparison of results of the proposed technique versus other deep learning models and machine learning techniques has been provided in the section VIII and the visualization of the results has been provided in section IX. The novelty of the method proposed is that it has memory to differentiate tuples with different order of scores and learn to assign the weights of relationship between nodes by scanning the sequence in both directions compared to decision tree, SVM, feed forward neural network based algorithms which have been earlier used to solve this problem of predicting student score.