Selection of Machine Learning Models for Prescription Decision-Making Based on Text Mining - Focusing on Case Studies of Single Prescriptions in Sasang Constitutional Medicine

Kim, Yunseo; Lee, Eunsu; Jeong, Roa; Park, Sohyun; Kim, Jinseok; Kim, Hyeonseo; Yu, Jun-sang; Kim, Yunseo; Lee, Eunsu; Jeong, Roa; Park, Sohyun; Kim, Jinseok; Kim, Hyeonseo; Yu, Jun-sang

doi:2025.46.1.70

JKM > Volume 46(1); 2025 > Article

ORIGINAL ARTICLE

J Korean Med. 2025;46(1): 70-86. doi: https://doi.org/10.13048/jkm.25006

텍스트 마이닝 기반 처방 결정을 위한 머신러닝 모델 선정 - 사상체질 단일 처방 치험례를 활용하여

김윤서¹

¹상지대학교 한의과대학
²서강대학교 국어국문학과
³상지대학교 한의과대학 사상체질의학교실
⁴상지대학교 한의학연구소

Selection of Machine Learning Models for Prescription Decision-Making Based on Text Mining - Focusing on Case Studies of Single Prescriptions in Sasang Constitutional Medicine

, and Jun-sang Yu^3,4

¹Department of Korean Medicine, Sangji University
²Department of Korean Language & Literature, Sogang University
³Department of Sasang Constitutional Medicine, College of Korean Medicine, Sangji University
⁴Research Institute of Korean Medicine, Sangji University

Corresponding Author: Jun-sang Yu ,Tel: +82-33-741-9203, Email: hiruok@sangji.ac.kr

Received: January 20, 2025; Revised: January 20, 2025. Accepted: February 11, 2025.

ABSTRACT

Objectives: : We analyzed Sasang constitution case reports using text mining and designed a classification algorithm using machine learning to select a model suitable for determining Sasang constitution prescriptions based on text data.
Methods: Case reports on Sasang constitution published from January 1, 2000, to December 31, 2023, were collected. A total of 360 papers and 483 cases were identified, from which text was extracted for 253 cases. The extracted texts were preprocessed and tokenized using the Python-based KoNLPy package, and each morpheme was vectorized using TF-IDF values. To select the most suitable classification model for diagnosing Sasang constitution, the performance of five models—Random Forest Classifier, XGBoost, LightGBM, SVM, and Logistic Regression—was evaluated based on accuracy and F1-Score.
Results: The highest accuracy was achieved by Random Forest Classifier (0.57037), followed by SVM (0.544444), Logistic Regression (0.518519), LightGBM (0.481481), and XGBoost (0.474074). The F1 score was highest for Random Forest Classifier (0.528), followed by SVM (0.52039), Logistic Regression (0.500861), XGBoost (0.45866), and LightGBM (0.446349).
Conclusions: This study is the first to analyze Sasang constitution prescription decisions by applying text mining and machine learning to case reports, providing a concrete research model for follow-up studies. Based on case reports and text data, the most suitable machine learning model for determining Sasang constitution prescriptions is Random Forest Classifier.

Keywords: Data Mining | Machine Learning | Case reports | Four-Constitution Medicine | Random Forest Classifier