Machine learning for tongue diagnosis

Tongue diagnosis is an important inspection method in Traditional Chinese Medicine (TCM). In this paper, we investigate machine learning techniques for tongue diagnosis. To do this, we first identify tongue properties and classes. In tongue property identification, we identify 21 properties from tongue substance and coating, whereas in tongue classification, we derive 24 tongue classes. Machine learning techniques are then applied to a tongue dataset. In performance analysis, we use the Weka machine learning environment for conducting the experiment. Five different machine learning algorithms including ID3, J48, Naive Bayes, BayesNet and SMO are used and applied to a tongue dataset of 457 instances. The performance results have shown that the Support Vector Machine algorithm SMO has the best performance for tongue diagnosis based on accuracy and Area Under the ROC Curve (AUC).