Paper Title :Academic Performance Prediction Using Artificial Neural Networks Based on Course-Related Student Data

Author :Hon-Sun Chiu, Zhi-Ting Billy Ng, Yuk-Lun Cheung

Article Citation :Hon-Sun Chiu ,Zhi-Ting Billy Ng ,Yuk-Lun Cheung , (2023 ) " Academic Performance Prediction Using Artificial Neural Networks Based on Course-Related Student Data " , International Journal of Advance Computational Engineering and Networking (IJACEN) , pp. 15-20, Volume-11,Issue-8

Abstract : In recent years, e-learning has become an essential implementation in education. A lot of e-learning systems have been developed to support online courses and hybrid face-to-face classes with online activities. These systems recorded a hugh amount of data related to students’ learning characteristics, such as the frequency of accessing learning materials, activeness in participating online activities, timeliness in assignment submission, and engagement in forum discussion. These data are considered highly correlated to students’ academic performance. If teachers can analyze the data and identify weak students at an earlier stage, necessary support can be provided to them that may improve their performance. In this paper, we developed four academic performance prediction models using deep artificial neural network (DNN). Different from other existing prediction models in the literature that involves students’ personal and privacy data, such as social status, home address and other family factors, our models only consider course-related data. Such implementation does not only protect the privacy of students, these data can also be assessible by normal course teachers directly. To study how early the student performance can be correctly predicted, the four models were implemented in different stages along a semester. Our results showed that the models can achieve the highest prediction accuracy of 91.30%, while the performance can be predicted as early as when the first assignment was collected, with the accuracy of 85.56% and recall value of 0.89. This also means that 89% of “should-be-failed” students can be identified correctly at a very early stage. Our findings could help educators in planning student supports to avoid withdrawal and underachievement. Keywords - E-learning, Artificial Intelligence, Neural Networks, Academic Performance Prediction.

Type : Research paper

Published : Volume-11,Issue-8

	|		PDF	|	Viewed - 24	|	Published on 2023-11-27