Abstract: Differentiating sarcoidosis from tuberculosis is still difficult．The support vector machine is a powerful tool in statistical learning.In this paper,we collected 106cases of sarcoidosis and tuberculosis,used an SVM to build a disease classifier named ADST(Automated Differentiating Sarcoidosis from Tuberculosis).In order to get the raw medical data into a form usable by SVM,we extracted feature vectors of the raw medical data by turnning the qualitative feature into digital one and dropping the features that do not have much classification value.Then ADST conducts simple scaling on the data,uses cross-validation to find the best parameter of model,uses the best parameter to train the whole training set to obtain the SVM model.Finally ADST uses the resulted SVM model to predict a new patient case.The experiment result shows that the ROC areas of SVM,DCT and NB are 0.978,0.96,0.690respectively,and the training accuracy is 95.28%,90.57%,92.38%,and test accuracy is 100%,96.15%,96.15%．Clinical pratice shows that the classification result is correct:19cases of undiagnosed patients are recovered after treatment according to the results of the diagnosis of ADST.

Key words: Sarcoidosis,Tuberculosis,Statistical learning,SVM