Tax compliance plays a pivotal role in optimizing state revenue to support regional development and public infrastructure. At the Sibolga Primary Tax Office (KPP Pratama Sibolga), tax administrators face operational challenges in manually identifying non-compliant taxpayers due to high administrative data volumes and complex taxpayer behavioral attributes. This study proposes an automated machine learning framework to classify taxpayer compliance levels based on multidimensional taxpayer awareness indicators using the Naïve Bayes algorithm. Primary empirical data was collected from 250 registered individual taxpayers (Wajib Pajak Orang Pribadi) under KPP Pratama Sibolga through structured questionnaires covering tax regulation understanding, penalty perception, service quality satisfaction, and filing consistency. The data was preprocessed, categorized, and evaluated using a 10-Fold Cross-Validation setup. The experimental results demonstrate that the Naïve Bayes model achieves an overall classification accuracy of 90.80%, with a precision of 88.30%, a recall of 87.37%, and an F1-score of 87.83% for predicting non-compliant taxpayers. Comparative analysis against Decision Tree (C4.5) and Support Vector Machine (SVM) models confirms that Naïve Bayes provides superior predictive performance and computational efficiency (12 ms execution time) for discrete survey-based datasets. The conditional probability findings reveal that low regulation awareness and rare filing consistency serve as the primary behavioral drivers of tax non-compliance. Practically, this probabilistic model provides KPP Pratama Sibolga with an evidence-based administrative tool to optimize compliance audit selection, deploy targeted tax socialization programs, and improve voluntary tax compliance
Copyrights © 2026