Estimation of ROC curves in the presence of measurement errors (KU LEUVEN)
Completed.
We concerned the problem of estimation receiver operating characteristic (ROC) curves in the presence of measurement errors. A ROC curve enunciates the probability of a true positive (PTP) as a function of the probability of a false positive (PFP) for all possible values of the cutoff between cases and controls. The area under the curve,
$\theta$ , can measure globally how well the separator variable distinguishes between cases and control. Therefore, the area under the curve,$\theta$ , is widely used as a summary measure of diagnostic accuracy. We propose a smooth non-parametric ROC curve derived from Bernstein type polynomial estimates to obtain the ROC curve and the area under the curve. The features of the Bernstein polynomial can take the noted drawbacks of non-parametric ROC curve in hands. The aim of the paper is the estimation of the ROC curve and the Area under the curve (AUC) when predictors are measured with error.
In Advanced Nonparametric Statistics and Smoothing course in KU LEUVEN, I wrote a report based on the paper, "Efficient and robust density estimation using
Bernstein type polynomials" written by Zhong Guan paper. You can find the report in the file, nonparametric_research_paper, under reference_file folder.