A Comparative Study of Student Performance Prediction using Pre-Course Data

Abstract

Students at Saudi universities face difficulty registering for the right course since Student performance there is no support offered to students that uniquely consider each situation. Machine learning techniques could be applied to fill this gap by predicting grades of new courses for each student based on their historical data. This paper experiments with nine different prediction algorithms to predict course grades for public university students. The data-set includes grades for 215 students and 180 various courses. The models utilize grades obtained in semesters between the 2015 and 2018 academic years and evaluated on grades obtained in the 2019 academic year. Our result shows that the K-nearest neighbor with ZScore model outperforms the remaining models with respect to the Percentage of Tick Accuracy (PTA), which is the difference between two consecutive letter grades for the predicted letter grade and the observed letter grade. Our work achieved an 84% accuracy score in PTA2, where the difference between the predicted letter grade and the actual letter grade is less than or equal to two consecutive letter grades.

Two-Way Neural Network Performance Prediction Model Based on Knowledge Evolution and Individual Similarity

“…More and more machine learning algorithms successfully applied in other fields are being introduced to education. Alharbi et al [37] explored using different prediction algorithms to predict college students' grades, such as K-Nearest Neighbor (KNN), Singular Value Decomposition (SVD), and Non-negative Matrix Factorization (NMF). Many literature [38][39][40] adopted Matrix Factorization (MF) to learn the embedding for each student and course and predicts the grades based on corresponding vector embeddings of the course and student.…”

“…And some studies are based on such traits to predict student performance. For instance, Alharbi et al [37] studied use K-Nearest Neighbor (KNN) algorithm to predict students' performance in the course. In recent years, graphical neural networks (GNN) [52] have been successfully applied in many fields, such as relationship mining, medical diagnosis, and personalized recommendation.…”

“…Table 2 shows the grade levels and their associated numerical score ranges. Predicting which level of students' course learning achievements can be viewed as an ordered category classification prediction issue, and its frequent evaluation measure is the Percentage of Tick Accuracy (PTA) [37]. For calculating PTA, the first step is to convert the ground truth score values, and the corresponding predicted score values to the grade level.…”