Implementation of a Hybrid Model Using Principal Component Analysis, K-Means, and Naïve Bayes for Tuition Fee Category Prediction

Abstract

The determination of Tuition Fee Categories in higher education institutions is commonly conducted through manual verification of students’ socioeconomic documents, which may lead to subjectivity and inconsistencies in decision-making. This study proposes a hybrid machine learning approach that integrates Principal Component Analysis (PCA), K-Means Clustering, and Naïve Bayes Classifier within a semi-supervised learning framework for student socioeconomic classification based on pseudo-labels generated from clustering results. The dataset used in this study consists of 452 student records with 12 socioeconomic attributes obtained from the New Student Admission system of STAIN Teungku Dirundeng Meulaboh in 2025. Data preprocessing includes attribute selection, categorical transformation using One Hot Encoding, and feature standardization. PCA is applied to reduce dimensionality from 18 features to 12 principal components while retaining 95% of the total variance. The processed data are clustered using K-Means with the optimal number of clusters determined as 8 based on Elbow and Silhouette Score analysis. These clusters are used as pseudo-labels for training the Naïve Bayes classifier. Experimental results show that the proposed model achieves 98.89% training accuracy and 97.80% testing accuracy, with a weighted average F1-score of 0.98. The results indicate that the proposed hybrid approach is effective in capturing underlying socioeconomic patterns and provides a stable classification performance. However, the model is based on pseudo-labels rather than official tuition fee categories. Therefore, further validation using real labeled data is recommended to enhance generalizability and practical applicability.