The Use of a New Classifier to Maximize the Classification Performance

• Activity Number: 339 - SPEED: Biopharmaceutical and General Health Studies: Statistical Methods and Applications, Part 1
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Hua Ma* and Joe Heyse
• Companies: Merck and Merck
• Keywords: ROC; distribution overlap; classification accuracy; Lanchenbruch's ?; machine learning; AUROC

Abstract

Statistical methods have been constantly developed for maximizing classification performance. The overall classification performance for a diagnostic test can be evaluated based on different metrics, e.g. the area under the ROC curve (AUROC), Lachenbruch’s ?, the measure of distribution overlap, and the overall classification accuracy. New classifiers are proposed by synthesizing densities based upon Fisher’s method and direct computation of multivariate density function. Under the commonly used criteria for evaluating overall classification accuracy mentioned above, the classification performance of the proposed methods will be compared using statistical simulation and real data examples with the current popular established methods, e.g. regression, and machine learning methods such as random forest and support-vector machine (SVM). Different simulation scenarios will be considered and implemented.