A new method for identification of traditional Chinese medicine constitution based on tongue features with machine learning.

This article focuses on developing an efficient and accurate model for identifying Traditional Chinese Medicine (TCM) constitutions using objective tongue image features combined with machine learning techniques. Utilizing 1,149 tongue images representing nine TCM constitution types, the study extracted 45 tongue features and applied five machine learning algorithms—Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), LightGBM (LGBM), and CatBoost (CB)—to build classification models. The authors constructed a heterogeneous ensemble learning model named RLC-Stacking, which integrated RF, LGBM, and CB as base learners and logistic regression as a meta-learner, achieving improved accuracy (up to 82.87%) and robust identification performance across all constitution types. Feature selection further enhanced model performance, and explainable machine learning techniques (SHAP) were used to interpret feature contributions. The study demonstrates that tongue image analysis combined with ensemble learning provides a reliable, objective, and rapid method for TCM constitution identification, supporting personalized healthcare and clinical decision-making.