EWMA model based shift-detection methods for detecting covariate shifts in non-stationary environments

作者：

Highlights：

• The main focus is on the covariate shift-detection tests based on EWMA.

• In univariate shift-detection test, getting the excessive false-alarms is an issue.

• The issue of false-alarms has been handled by a novel two-stage structure test.

• A multivariate formulation for the covariate shift-detection is also presented.

• The proposed methods are superior in accuracy and reducing the false-alarms.

摘要

•The main focus is on the covariate shift-detection tests based on EWMA.•In univariate shift-detection test, getting the excessive false-alarms is an issue.•The issue of false-alarms has been handled by a novel two-stage structure test.•A multivariate formulation for the covariate shift-detection is also presented.•The proposed methods are superior in accuracy and reducing the false-alarms.

论文关键词：Non-stationary environments,Dataset shift-detection,Covariate shift,EWMA

论文评审过程：Received 18 December 2013, Revised 20 June 2014, Accepted 26 July 2014, Available online 5 August 2014.

论文官网地址：https://doi.org/10.1016/j.patcog.2014.07.028