Machine Learning-Based Cyber Surveillance System for Detecting Illegal Pharmaceutical Product Distribution

Abstract

The rapid development of digital technology has significantly transformed pharmaceutical distribution patterns, including the increasing circulation of illegal products through social media and e-commerce platforms. This condition complicates post-market surveillance due to the fast, massive, and difficult-to-detect nature of digital distribution using conventional monitoring approaches. This study aims to develop a machine learning-based cyber surveillance system for detecting illegal pharmaceutical product distribution and mapping spatial patterns of violations in Papua and Papua Pegunungan regions. A quantitative experimental approach was employed using cyber patrol data from the first quarter of 2026 consisting of 162 violation records. The classification models used were Support Vector Machine (SVM) and Random Forest (RF) with data splitting scenarios of 70:30, 80:20, and 90:10, evaluated using accuracy, precision, recall, weighted F1-score, and macro F1-score. The results show that Random Forest consistently outperforms Support Vector Machine, achieving accuracy ranging from 93.88% to 100%, while SVM achieved accuracy between 87.76% and 94.12% across all testing scenarios. Furthermore, 3-Fold Cross Validation indicates that Random Forest achieves a higher average accuracy of 89.51% with a lower standard deviation of 0.0231 compared to SVM, which records an average accuracy of 83.33% with a standard deviation of 0.0400, indicating better stability and generalization capability. Spatial analysis identifies Jayapura City as the region with the highest concentration of violations, supporting a risk-based surveillance approach. The integration of machine learning and spatial analysis is proven to enhance detection accuracy, model stability, and data-driven decision-making in digital pharmaceutical surveillance.