NOVEL STUDY OF DEEP LEARNING ALGORITHMS FOR STUDENTS' PERFORMANCE PREDICTION

Abstract

Higher education refers to that phase of education that starts after schooling. Universities offer graduate courses in multiple disciplines, and students select courses that interest them. Both the government and private sectors run universities in India. These institutions certify the students based on their performance in the academic sessions. These universities compete with each other in providing quality education that opens new avenues for further studies or employment globally. Due to the intense competition in the education and employment sectors, designing a robust assessment system for students' performance has become imperative. Given the vast majority of the student population and their potential in various domains, there is a need for a computer-aided algorithm that can analyse their performance with the highest possible degree of accuracy. The existing methods that used machine learning algorithms to assess student performance gave less accuracy and could not handle massive datasets. This paper presents a deep learning methodology capable of managing diverse datasets, ranging from minimal to extensive, incorporating a Deep Artificial Neural Network (DANN), a Convolutional Neural Network (CNN), and Bidirectional Long Short-Term Memory (BiLSTM) algorithms, along with hybrid deep learning architectures, including DANN-BiLSTM and CNN-BiLSTM. This approach aims to assess student performance, known as the student performance grading system.

The research shows that the proposed Students Performance Grading system gave accurate results on various factors of training and testing compared to the existing ones. The results show that the MSE of CNN is 0.0227, the MSE of Deep ANN is 0.0219, and the MSE of Bi-LSTM is 0.0233. Additionally, the MSEs are 0.0211 for DANN-Bi-LSTM and 0.0215 for CNN-BiLSTM. The accuracies are 96.74% for DANN, 96.42% for CNN, and 95.80% for Bi-LSTM. and DANN-Bi-LSTM attained an accuracy of 99.31%, while CNN-Bi-LSTM achieved an accuracy of 97.30%. This research identified that DANN-Bi-LSTM is the most effective algorithm for forecasting student success.