COMPARATIVE STUDY OF CLASSIFICATION MODELS IN PROCESSING STUDENT TEST SCORES DATASETS

Abstract

The development of Machine Learning (ML) has contributed significantly to the field of education, particularly in analyzing student academic data to support data-driven decision-making. Predicting student exam results is important for identifying academic performance patterns, detecting potential failures, and improving learning interventions. However, variations in student characteristics and dataset complexity require the selection of appropriate classification models to achieve optimal prediction performance. This study aims to compare the effectiveness of several ML classification models in predicting student exam results using a student academic dataset. The dataset consists of 306 records, seven attributes, and five grade classes (A, B, C, D, and E), including attendance, quiz scores, midterm examination scores, final examination scores, and assignment scores. Data preprocessing was conducted to handle missing values, duplication, inconsistencies, and outliers. The dataset was split into training and testing data with a ratio of 75:25 and evaluated using 10-fold cross-validation. Several classification models were applied, including k-Nearest Neighbour (kNN), Decision Tree, Naive Bayes, Support Vector Machine (SVM), and Random Forest. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics. The experimental results showed that Random Forest achieved the best performance with an accuracy of 73.9%, precision of 74.0%, recall of 73.9%, and F1-score of 73.9%, followed by Naive Bayes and Decision Tree. Meanwhile, SVM produced the lowest performance among the tested models. The findings indicate that Random Forest is the most effective method for predicting student exam results and has strong potential to support educational decision-making systems.