AI-Based Decision Support for Traditional Medicine Governance: Classifying Jamu Efficacy Using Random Forest

Abstract

Jamu is an important part of Indonesian traditional medicine, but information about its ingredients and reported benefits is often scattered and difficult to organize systematically. This condition creates challenges not only for consumers, but also for public institutions that need reliable data to support education, supervision, and traditional medicine governance. This study developed a Random Forest classification model to classify reported jamu efficacy based on dominant ingredients, ingredient percentage, and stated health benefits. The dataset consisted of 205 commercial jamu products collected from publicly available sources. Text variables were cleaned and transformed using TF-IDF vectorization, while class imbalance was handled using random oversampling. The model was evaluated using repeated 10-fold cross validation with accuracy, macro precision, macro recall, and macro F1-score as performance metrics. The results showed that Random Forest achieved an accuracy of approximately 0.897, with balanced macro precision, recall, and F1-score. The findings suggest that machine learning can support the early organization of traditional medicine information, although it cannot replace clinical or regulatory validation.