Modeling the performance of an academic course based on a given set of affecting factors is the goal of this research. For different institutions, these factors differ in terms of availability and usefulness. This study was conducted for the nine engineering departments at King Abdulaziz University, Saudi Arabia with a total of 281 courses for the last 8 years. First, all measurable input factors were acquired from the database, and a comprehensive statistical study to course performance was performed. In modeling the input factors to the course performance, an adaptive linear model was first implemented at three levels: the college level, the department level, and the course level. Results show that the linear model fitted only 49% of the courses with an error standard deviation of 5.41 grade points, which is above the target of 2.5. On the other hand, the proposed neural network model showed much promising results: 83% of the courses were fitted with an error standard deviation of 0.96, having 95.26% of courses being modeled perfectly. In regard to the neural network structure and type, an exhaustive analysis was conducted by constructing and training 71,295 neural networks. It showed that the feed-forward and the cascade-forward types are the best with hidden layers between two to three.
Key words: Course performance, modeling, neural network, performance indicators, statistical testing.
Copyright © 2021 Author(s) retain the copyright of this article.
This article is published under the terms of the Creative Commons Attribution License 4.0