A Multi-class Classification Method Based On ROC Analysis

Accurately identify the true status of tumors is crucial for the following treatment. But some of certain tumors are quite similar, or their early symptom is not obvious enough in diagnosis. These tumors can hardly be diagnosed in an early stage using traditional methods.In this paper, we propose a classification and feature selection method for multi-class tumor classification, and apply it in 2 tumor data. We sup-pose in each class, samples are of normal distribution, and the features are independent of each other. We build Bayes posterior probability classifier with estimated parameter, and classify each sample to the class with high-est probability. Then we use HUM to evaluate the classification effect and select feature gene. We use SRBCT and 14-cancer data as examples of its application. The results indicate that our method is comparatively efficient in tumor classification problems.