Research on the eight classifications of traditional Chinese medicine biased constitution based on random forest feature selection and SMOTE+ENN algorithm

Abstract:
Introduction Currently, the main method of identifying biased body types is to use self-assessment and scoring scales for classification, which is not only influenced by the subjective perception of the self-assessor but also takes a lot of time. Based on this, this paper proposes to classify biased body types based on acupuncture point signals. Research methods In this study, 1562 subjects were recruited to conduct the questionnaire named " Traditional Chinese Medicine (TCM) Constitution and Classification Self-Test Table" of professor Wang Qi with the method of cross-sectional survey and the scale score was considered as the standard to determine the type of constitution. In addition, signals of 24 original points on 12 meridians and basic information including height, weight, age and gender were collected and constructed as database. Then, a random forest-based feature selection method was used to select relevant features from the original dataset to form the optimal feature subset. To address the problem of unbalanced dataset categories, the SMOTE+ENN algorithm is used for balancing, and the dataset after feature selection and balancing is trained to obtain a biased eight-level classification model. Results The biased body eight classification model was trained 100 times after, the mean of accuracy, mean of sensitivity, mean of specificity, and mean of F1 score were 75.6%, 75.03%, 75.20% and 74.79% respectively. Conclusion A classification model based on random forest feature selection and the SMOTE+ENN algorithm is instructive for the classification of biased Constitution.