Disease Analysis Based On Hierarchical Classification

The cures of TCM (Traditional Chinese Medicine) experts are crystallization of famous doctors’ experience, they contain abundant knowledge and experience. So the inheritance has great significance, and the data mining of TCM is an important work. The article selects the records of vertigo, which has high incidence. It has a trend of getting younger over the onset age, and the condition will break out repeatedly, which interferes with the normal life and work, so researching the causes of vertigo is necessary. The paper researches the records through some data mining methods from two aspects, firstly analyze based on the hierarchical classification method from the symptoms, to support the result, we use the clustering method based on active learning from the prescription. The main content of this article is as follows:1. Based on the k-means which divides the samples to two partitions, we design an improved relax hierarchical method, which is suitable for most of the top-down hierarchy. When there is uncertainty in decision-making, the method suggests the classes can be divided into overlap partitions, so the decision can delay until the number of categories decrease, so it can easily learn good decision boundary. Then based on the SVM classifier, we design a kind of classification method, which can classify fast when there are many categories. A set of binary classifiers is organized into a DAG structure, which is developed under the label space hierarchy. The key to this method is the color of classes, and the coupling study of class colors and binary classifiers with Max-margin optimization, which has better performance.2. Aiming at the problem that k-means can not determine the initial clustering centers, resulting in the low accuracy, we design a clustering algorithm by applying Min-Max active learning strategy to ask the user to identify the seed points, which improves the accuracy of results in prescriptions. We analyze the microblog with the method, which has a good result.3. The above hierarchical classification method is applied to vertigo symptoms, which can be divided into five types of etiology, and it conforms to the theoretical knowledge of TCM books; also we apply clustering analysis based on active learning to the prescription of vertigo, which can verify the classification results, and has a high accuracy.4. Based on the developed functions on the early TCM data mining platform and the above optimization algorithm, we extend the function of the platform.