Deep Learning approach for predicting the therapeutic usage of Jamu

Abstract

As a country with the largest number of medicinal plants in the world, Indonesian uses medicinal plants as a composition of herbal medicine. The ingredient of herbal medicine is generally made based on experiences and hereditary. This research aims to build a scientific background of Jamu through analysis of the relationship between medicinal plants used as the composition of Jamu and its therapeutic usage. Deep Learning was chosen as a classifier because it shows good effectiveness in generating predictive models in many studies. As a comparison, we also applied the Random Forest and Support Vector Machine as classifiers and examined the classifier performances while predicting the therapeutic usage of Jamu. To handle the imbalanced data between efficacy classes, the Synthetic Minority Oversampling Technique was applied before model generation. The result shows that the highest accuracy for Deep Learning is 88.74%, relatively higher than Random Forest and Support Vector Machine, which obtain accuracy values of 78.84% and 77.60%, respectively. Variable importance of the best prediction model using Deep Learning identified 105 medicinal plants, and 39 of them were selected as potential plants for 14 therapeutic usages.