Research Article: A multi hidden recurrent neural network with a modified grey wolf optimizer

Date Published: March 27, 2019

Publisher: Public Library of Science

Author(s): Tarik A. Rashid, Dosti K. Abbas, Yalin K. Turel, A Lenin Fred.

http://doi.org/10.1371/journal.pone.0213237

Abstract

Identifying university students’ weaknesses results in better learning and can function as an early warning system to enable students to improve. However, the satisfaction level of existing systems is not promising. New and dynamic hybrid systems are needed to imitate this mechanism. A hybrid system (a modified Recurrent Neural Network with an adapted Grey Wolf Optimizer) is used to forecast students’ outcomes. This proposed system would improve instruction by the faculty and enhance the students’ learning experiences. The results show that a modified recurrent neural network with an adapted Grey Wolf Optimizer has the best accuracy when compared with other models.

Partial Text

In education management, student performance prediction and classification systems are important tools. They warn students who did not perform well or those with at risk performance and assist students in averting and overcoming most of the problems they face in meeting their objectives. Yet, there are challenges in gauging students’ performance, since academic performance depends on various elements, such as demographics, personalities, education background, psychological issues, academic progress and other environmental variables [1].

In this section, the related works of two concepts are discussed in two parts, as follows: the state of the art applications for forecasting student performance and the state of the art grey wolf optimizer applications with/without neural networks.

In this research work, a Grey Wolf Optimizer algorithm was modified. Then, this modified version was applied for optimizing the weight and bias of a modified recurrent neural network to predict student performance. Details about both the standard Grey Wolf Optimizer and Recurrent Neural Networks are first explained.

A modified recurrent neural network with a modified GWO was used for predicting student performance. This research improves on the previous study on student academic performance in [32]. The problems of back-propagation have been highlighted and the data have been collected from our previous research work about student performance in English courses at the College of Engineering at Salahaddin University. The data consist of 287 samples. In this proposed approach, an RNN model is developed by using the modified GWO to optimize the values of biases and weights of the model. Initially, the neural network model is trained by using a training dataset, and its weights and biases are optimized by using a modified recurrent network with GWO. In the second step, to evaluate the trained model, the designed model is tested with a predefined testing dataset. For the validation procedure, cross validation of 5-fold is used for attaining high accuracy and performance. In this study, MATLAB is used for the implementation. The key stages of this work are explained below:

The results of the classification using cross validation are shown in Table 1. The dataset was divided into five groups (5-folds), named as X1, X2, X3, X4, and X5. The first three groups consisted of 57 samples, and the last two group contained 58 samples. In each fold run, four groups were fed to the network model, as the training dataset consisted of approximately 230 samples, and the remaining were rolled, as the testing dataset consisted of approximately 57 samples to test the network. The results showed that the training classification rates in the folds were 99.56%, 99.56%, 99.56%, 99.12%, and 99.56%, and the average rate was 99.47%. Also, the classification rates for the testing phase for each fold were 96.49%, 100%, 100%, 98.27%, and 98.27%, and the average was 98.60%. It can be seen from the results that when a smaller TotalMSE is produced, a better classification rate is obtained. For example, in Fold 1, the classification rate in the testing phase is 96.49% and its TotalMSE  is 0.009, but when the testing rate is 100.00% in the second and third folds, then the total MSE is 0.002, which is a smaller MSE.

In this paper, a student performance system was suggested for classifying students in English courses based on their previous accomplishments, social setting, and academic setting. The classification technique used a modified GWO for optimization of weights and biases of a modified RNN model. The modification in the GWO involved inserting another best solution into the population of the wolves. Also, the average of the distance of the best wolves was taken into consideration instead of taking the separate distances of the best wolves. This modification had a good effect, since the position of the search agents was updated with an extra best solution. The concept involved the simple RNN type based on an MLP with two hidden layers as a classifier for the prediction of student outcomes. In general, the aim of using meta-heuristic methods with a neural network is to maximize the outcome of the neural network model. The results demonstrated that the proposed adaptation enhanced the students’ performance positively.

 

Source:

http://doi.org/10.1371/journal.pone.0213237

 

Leave a Reply

Your email address will not be published.