Analytical study of performance of linear discriminant analysis in stochastic settings
作者：
Highlights：
• We consider the true error of linear discriminant analysis (LDA) for non-i.i.d data.
• The first and second moments of true error for univariate LDA are characterized.
• The classification performance (CP) for non-i.i.d. data is compared to i.i.d. setting.
• As an application, we consider auto-regressive (AR) or moving-average (MA) models.
• We find ranges of AR/MA coefficients that have incremental or decremental effect on CP.
摘要
Highlights•We consider the true error of linear discriminant analysis (LDA) for non-i.i.d data.•The first and second moments of true error for univariate LDA are characterized.•The classification performance (CP) for non-i.i.d. data is compared to i.i.d. setting.•As an application, we consider auto-regressive (AR) or moving-average (MA) models.•We find ranges of AR/MA coefficients that have incremental or decremental effect on CP.
论文关键词：Linear discriminant analysis,Stochastic settings,Correlated data,Non-i.i.d data,Expected error,Gaussian processes,Auto-regressive models,Moving-average models
论文评审过程：Received 28 November 2012, Revised 3 April 2013, Accepted 7 April 2013, Available online 16 April 2013.
论文官网地址：https://doi.org/10.1016/j.patcog.2013.04.002
