Data Mining About Syndrome For Traditional Chinese Medicine

Traditional Chinese medicine (TCM), one of the most important complementary and alternative medicines, is invaluable for its rich practical knowledge and a unique integrated theoretical system established since ancient times. TCM has been clinically observed to have dramatic performance in treating many chronic and systematic diseases such as the treatment of liver cirrhosis. During the diagnosis, syndrome differentiation is the most key step. Syndrome differentiation is the method of recognizing and diagnosing diseases or body imbalances by analyzing patient information based on TCM theories and the doctor's experiences. Syndrome enables the doctor to determine the stage that the disease developed and the location of the disease. However, the lack of objective diagnosis standards hinders TCM wide acceptance. One cannot apply this prescriptive methodology in a professional standard until or unless one has mastered the syndrome differentiation process. The purpose of this paper is in an attempt to achieve effective and objective standard of syndrome prediction.In order to obtain the objective rule of syndrome differentiation, this paper applies the data mining technique into TCM syndrome prediction.The inherently nonlinear, ambiguous and complex characteristic of TCM data increases the difficulty of data mining for TCM. The research on syndrome differentiation can't be the simple cause and effect between the symptoms and the syndrome. On the basis of understanding and analyzing the current syndrome differentiation research state and data mining relating algorithm, the main achievement of this paper is as follows:1. Feature selection based on both the TCM view and Western medicine view The dataset of TCM has objective and subjective features, the number is huge. At the same time, collecting data is never an easy job in liver cirrhosis applications because of time consuming and costly work. Feature selection is the key step in the data mining of TCM. Traditional Chinese medicine and Western medicine unite in essence and should be combined to predict diagnosis. Based on the TCM view and the Western medicine view, a noel method called BVFS (Bi-View Feature Selection) is proposed. Instead of only using TCM symptoms and signs for syndrome prediction, raw dataset contains both TCM data and Western medicine data. Therefore, the critical features are more objective while it retains the subjective part to some extent.2. Attribute hierarchy syndrome differentiation model to the lack syndrome differentiation ruleIn the most syndrome differentiation system, the rule of syndrome differentiation is not standard and need some experiences of experts. In this paper, based on some new concepts such as the combination attribute-measure, discrete attribute-measure, a novel method AHSDM (attribute hierarchy syndrome differentiation model) is proposed, which distinguishes the system from those existing TCM diagnosis systems based on classification algorithm to address syndrome classification. This model absorbs the advantages of subjective attribute-measure from expert's experience and objective attribute-measure from original cases to make medical decision. Then subjective and objective attribute measures will be combined to form the combination attribute-measure from which we get the optimal solution.3. Ensemble syndrome differentiation model to the full syndrome differentiation ruleSome of the syndrome differentiation system, the standard rule of syndrome differentiation is built. During the constructing of syndrome differentiation model, single classifier can not improve the classification performance because of the difficulty TCM data. A new ensemble syndrome learning method based on the multi views is proposed, which trains the best classifiers in a local feature view. Based on these best classifiers, we apply the weighted-voting to determine the syndrome of a new case.4. Construction of the syndrome differentiation systemWe integrate some of the algorithms proposed in this paper to develop a syndrome differentiation system. In this system, we can obtain the optimal feature subset and predict the syndrome of a new case.