Explainable Prediction of Alternative Medicine Outcome using Machine Learning and Shapley Values

Alternative medicine is widely used by the public since more affordable and has minim side effects. Even so, several studies of acupuncture therapy have shown positive results but not in all patients. So that further research is needed regarding the effectiveness of alternative medicine to increase the prevalence of patient recovery. In this paper, we present a framework to predict the effectiveness of alternative medicine such as acupuncture, cupping, and the use of herbs. Dataset from medical records of patient alternative medicine at a health center in the Tangerang area within ten years. The best framework is realized in the XGBoost Classifier model with an accuracy value of 94.7% and a kappa score of 0.89. Shapley values identify the feature that contributed to the classification decision with the XGBoost Classifier model, demonstrating the high impact of inputs model is the number of therapies, duration of illness, type of disease, and duration of therapy.