Optimizing Machine Learning for Digital Surveillance and Pharmaceutical Product Risk Assessment

Abstract

This study explores the optimization of Machine Learning for digital surveillance and risk assessment of pharmaceutical products in Eastern Indonesia, a region with limited physical surveillance. The system used a Support Vector Machine (SVM) optimized with TF-IDF N-gram (1,2) and SMOTE to classify 3,539 pharmaceutical product advertisements into low, medium, and high risk. The results showed an accuracy of 98.47%, the ability to detect high risks in low-volume areas, as well as the identification of Jayapura-Nabire as the main distribution center and Southeast Sulawesi as a risk hotspot. Automatic screening of 1,000 ads takes only 25 minutes, compared to manual monitoring, which is much slower. The findings support proactive surveillance, precision risk classification, and strategic resource allocation in digital pharmaceutical surveillance.