Subset selection using Combined Analytical Signal
Alexey L. Pomerantsev, Oxana Ye. Rodionova
Abstract
We propose a new method of subset selection based on the concept of Combined Analytical Signal. It is equally suitable for selecting both a representative subset and a test subset, for single-block (classification) and multi-block (regression) data. As a result of simple calculations, each sample from the entire data set receives an importance index (SI) that takes values in the interval (0, 1), with larger values corresponding to more influential samples. This allows informed rather than blind subset selection. The first example shows that the advocated method has efficiency that is better than that obtained using the Kennard-Stone and D-optimal design methods. Another example provides the detailed guidance on splitting dataset into training and test subsets for a single-block data.