Abstract:

Multivariate imbalanced data arise from a variety of real world problems including rare disease studies, drug development,  electrical businesses and so on. Current goodness of fit tests are usually powerless to differentiate them. In this talk, we  introduce a local distance rank test procedure to detect the distribution equality of multivariate random variables in arbitrary  metric spaces. We show that the test statistic is consistent and converges to a mixture of χ2 distribution under null hypothesis  and some normal distributions under alternative hypothesis. Most important of all, we prove its consistency against alternative  hypothesis. This result does not depend on the ratio of two imbalanced sample sizes, which ensures that the test can work for  imbalanced data well. Numerical studies confirm that our test has better performance than several existing tests in view of Type  I error and power. We conclude this talk with applications of our method in a dataset of Virtual screening in drug development  process. 　　

报告人介绍：

王学钦，现为中山大学数学与计算科学学院和中山医学院双聘教授, 博士生导师，优秀青年研究基金获得者，兼任中山大学华南统计科学研究中心执行主任和数学与计算科学学院院长助理。主要研究方向为生物统计学、生物信息学和统计学习。在Nature Genetics、JASA、Biometrics、Genetic Epidemiology、PLOS one等等国际学术期刊上发表SCI索引论文30余篇。主持和参与多项国家和省部研究课题。此外，还担任教育部高等学校统计学类专业教学指导委员会委员、广东省统计学会副理事长、教育部高等学校统计学类专业教学指导委员会委员, 广东省第十届广东省青年联合会委员，广东省“领航100”计划青年科技特派员。