Searching for suitable classification methods in the discrimination of cold/heat herbal nature with Weka

Recent years have witnessed a rapid growth of data mining approaches in studies of Traditional Chinese Medicine (TCM), especially the medical theories. Using Weka workbench, we constructed seven representative classification models to discriminate the cold/heat nature of Chinese Herbal Medicine (CHM) based on efficacy attributes and evaluated them by three kinds of validations. The results showed that naive Bayes, Bayesian network and logistic regression performed efficient discrimination on the unseen test set after a proper attribute selection, while random tree, multi-layer perceptron network, logistic regression and PART algorithm might be suitable for the retrospective analysis with all attributes. This work could provide a methodology reference for further discrimination researches and facilitate the scientific explanation of these CHM concepts.