Enhancing Traditional Chinese Medicine Diagnosis through Machine Learning and Multidimensional Feature Integration

Abstract

Traditional Chinese Medicine (TCM) pattern recognition is the core component of traditional Chinese medical diagnosis. Integrating  modern machine learning technologies can enhance the objectivity and accuracy of diagnosis. This study constructed a multidimensional  feature system including demographic characteristics, medical history information, lifestyle factors, tongue image features, and  constitution types based on a balanced dataset of 1000 samples. Machine learning algorithms including Random Forest, Support Vector Machine, and Logistic Regression were employed to establish TCM pattern positive prediction models and analyze the correlations  between tongue image features and pattern modes in depth. Experimental results demonstrate that the Random Forest algorithm  performed optimally in pattern recognition tasks, achieving an accuracy of 0.873, F1-score of 0.869, and AUC value as high as 0.970. Feature correlation analysis reveals that constitution type has the strongest correlation with pattern positivity (r=0.72), while tongue  image features such as tongue color and coating thickness also demonstrate significant predictive value. Constitution type analysis  reveals that phlegm-dampness constitution patients have the highest pattern positive rate (91.2%), followed by damp-heat constitution  (88.7%) and blood-stasis constitution (84.5%), while balanced constitution patients have a positive rate of only 15.6%. This study provides theoretical foundation and technical support for the development of intelligent TCM diagnostic systems, promoting the deep  integration of traditional Chinese medicine with modern information technology.