接收机工作特性曲线研究-控制科学与工程专业论文.docxVIP

广东工业大学硕士学位论文ABSTRACT 广东工业大学硕士学位论文 ABSTRACT Receiver operating characteristic(ROC)analysis is important methodology to tackle the two—sample problems in many scientific and engineering fields．In essence， ROC analysis is supervised research framework that requires prior knowledge of both the sample membership and underlying cumulative distribution functions generating the two samples．Given such knowledge，the procedure of ROC analysis might be initiated from plotting two-dimensional ROC curve，which is traced out by pairs of false．．positive rate and true-·positive rate according to various decision threshold settings With the ROC curve，AUC can then be calculated，either analytically or empirically，as a figure of merit to assess how well a statistical model(also known as classifier in machine learning，diagnostic test in medicine or detector in signal processing)separates the group being tested into positive or negative categories in question．For circumstances with more than two alternatives，ROC and AUC have been extended in parallel to ROC surface and volume under the surface(VUS) Taking mathematical analysis and mathematical statistics as tools，in this paper we ＼＼ develop a complete framework for ROC a’nalysis in both parametric and nonparametric ways．The main contributions of this paper be summarized as follows 1)Based on an equivalent relationship between the Heaviside function and mid．ranks of samples，we improved DeLong’S algorithm in the sense of reducing the time complexity from the original quadratic order down to the present linearithmic order． 2)Starting from deriving closed-form expressions of the mean and variance of AUC for discrete samples，we compare four popularly used algorithms for estimating the variance of AUC in terms of unbiasedness and time complexity based on both artificial and real．world datasets 3)Develop recursive algorithm for computing the exact null distribution of the volume under the ordered three—class receiver operating characteristic surf