COMPARATIVE ANALYSIS OF MACHINE LEARNING ALGORITHMS FOR STUDENT PERFORMANCE PREDICTION
Keywords:
K-Nearest Neighbor, Educational Data Mining, Hyperparameter Optimization, Support Vector Machine, Decision Tree, Naïve Bayes, Machine Learning, Student Performance Prediction, K-means ClusteringAbstract
For university administrators, forecasting students’ academic performance is a major challenge. The majority of recent research examines several prediction “models” independently and excludes components like unsupervised feature selection and hyperparameter tuning, which degrades the “models’” quality and comparability. This study serves to fill in an existing research gap by systematically measuring and comparing four popular predictive “models” that were implemented under identical data cleaning and optimization methods Support Vector Machine (SVM), Decision Tree (DT), K-Nearest Neighbor (KNN), and Naive Bayes (NB). Those four predictive "models" were evaluated using a dataset of 32,005 students from Wollo University and Kombolcha Institute of Technology between 2017-2022. The goal was to evaluate the predicted student performance based on a K-Means algorithm (6 clusters) that took six different factors into consideration: gender, geographic location, university entrance exam scores, number of times the student has attempted the course, number of credit hours the student has completed, and the student's GPA in previous semesters. A total of 3,820 cumulative hyperparameter tuning iterations were carried out to optimize the four algorithms employed in this study in order to produce a model that would offer optimal performance accuracy. Subsequently, the dataset was analyzed using a 10-fold repeated cross validation. SVM showed the highest accuracy (96.0%) among the four classifiers used; followed by Decision Trees (93.4%), KNN (87.4%) and Naive Bayes (83.3%). According to per-grade-class performance statistics from grade A distinction, grade B failure and grade C pass , SVM would also be the best predictive analysis method for predicting a student’s academic success. Schools can use this information to create an early student identification program to identify students at risk of failing academically using predictive analysis.
