<![CDATA[This study evaluates and compares the performance of Naive Bayes and K-Nearest Neighbor (KNN) algorithms for classifying promotional-source categories in higher education admissions based on ten years of historical admission records. The objective is to analyze the capability of machine learning approaches in identifying patterns of applicant acquisition sources and to provide insights for institutional data-driven evaluation. After data preprocessing and quality filtering, 2,618 out of 4,901 records with complete target-variable information were retained and classified into seven promotional-source categories. Both algorithms were assessed using 5-fold cross-validation with multiple evaluation measures, including accuracy, macro-averaged recall, and comparison against a majority-class baseline to address the effect of severe class imbalance. Experimental results indicate that KNN achieved substantially higher overall accuracy (82.24%) than Naive Bayes (43.74%). However, neither model surpassed the majority-class baseline, demonstrating that accuracy alone can lead to misleading conclusions in highly imbalanced classification problems. In contrast, Naive Bayes obtained higher macro recall (36.32% compared with 18.14% for KNN), indicating a broader capability in recognizing minority promotional-source categories. The findings emphasize the importance of imbalance-aware evaluation and provide analytical insights into historical promotional-source distributions to support strategic admission planning and future institutional decision-making.]]>
Copyrights © 2026
