Abstract:

Objective    To explore the correlation between properties of Chinese traditional medicines（TCM）and their original characters, and to identify the feasibility of an error back-propagation artificial neural network (BP-ANN) to discriminate properties of TCM based on original characters.  Methods    Information on original characters and properties of 1728 kinds of TCM was first collected from  “Chinese Herbal Medicine”. 24 kinds of statistically significant characters were screened by single-factor and multi-factor logistic regression analysis. BP-ANN was applied to construct a model and to discriminate properties of TCM based on statistically significant characters.  Results    According to the principle of stratified random sampling, 60% of TCM in the cold group and the hot group were respectively selected as the training set and the remaining 40%(691 kinds in all) made up the testing set. A three-layer BP-ANN model was built to describe the correlation between properties of TCM and statistically significant characters; according to this model，the discriminative accuracy was 70.72% and 71.96% in the hot group and the cold group, respectively, and the overall accuracy was 71.49%.  Conclusion    Based on original characters, BP-ANN is capable of rapidly discriminating properties of TCM; BP-ANN, characteristic of adaptability, fault-tolerance and nonlinearity, and can effectively solve the problem of nonlinear correlation between properties of TCM and their original characters, providing a new idea and method for TCM property discrimination.

Key words: Artificial neural Network；  Property of Chinese traditional medicines； Discriminatory analysis