CMStatistics 2015: Start Registration
View Submission - CMStatistics
B1090
Title: Modeling the receiver operating characteristic curve using Gaussian and non-Gaussian mixture models Authors:  Amay Cheam - McMaster University (Canada) [presenting]
Paul McNicholas - McMaster University (Canada)
Abstract: The receiver operating characteristic (ROC) curve remains a topic of discussion and interest after all these years. The curve displays the capacity of a diagnostic test to distinguish between two groups of patients, diseased and non-diseased. In the literature, many approaches have been proposed for modeling the ROC curve whether it is direct or indirect. Because of its tractability, the Gaussian distribution has been extensively used to model both groups. In parallel, the finite mixture models have gained fame as a compelling apparatus in modeling data. We propose to model the ROC curve using Gaussian and non-Gaussian mixture distributions (specifically mixture of $t$ and skew $t$-distributions), leading to a more flexible model that accounts for heterogeneous data, unlike the classical binormal curve. The Monte Carlo method is used in conjunction to circumvent the absence of a closed-form and to obtain confidence bands for the derived ROC. The proposed method will be illustrated via simulated and real data.