Linear Decompositions for Multi-Valued Input Classification Functions

Abstract

In a multi-valued input classification function, each input combination represents properties of an object, while the output represents the class of the object. Each variable may have different radix. In most cases, the functions are partially defined. To represent multi-valued variables, both one-hot and minimum-length encoding are considered. Experimental results using University of California Irvine (UCI) benchmark functions show that the one-hot approach results in fewer variables than the minimum-length approach with linear decompositions.