Classification by Principal Component Regression in the Real and Hypercomplex Domains

Abstract Linear regression is a simple and widely used machine learning algorithm. It statistical approach for modeling the relationship between scalar variable one or more variables. In this paper, classification by principal component (CbPCR) strategy proposed. This depends on performing of each data class in terms its components. CbPCR formulation leads to new Regression Classification (LRC) problem that preserves key information classes while providing compact closed-form solutions. For sake image classification, also extended 4D hypercomplex domains take into account color image. Quaternion reduced biquaternion strategies are proposed representing channel as imaginary parts quaternion number. Experiments two face recognition benchmark databases show methods achieve better accuracies margin about 3% over original LRC like methods.