Study of the Class and Structural Changes Caused By Incorporating the Target Class Guided Feature Subsetting in High Dimensional Data

High dimensional data when processed by using various machine learning and pattern recognition techniques, it undergoes several changes. Dimensionality reduction is one such successfully used pre-processing technique to analyze and represent the high dimensional data that causes several structural changes to occur in the data through the process. The high-dimensional data when used to extract just the target class from among several classes that are spatially scattered then the philosophy of the dimensionality reduction is to find an optimal subset of features either from the original space or from the transformed space using the control set of the target class and then project the input space onto this optimal feature subspace. This paper is an exploratory analysis carried out to study the class properties and the structural properties that are affected due to the target class guided feature subsetting in specific. K-nearest neighbors and minimum spanning tree are employed to study the structural properties, and cluster analysis is applied to understand the target class and other class properties. The experimentation is conducted on the target class derived features on the selected bench mark data sets namely IRIS, AVIRIS Indiana Pine and ROSIS Pavia University data set. Experimentation is also extended to data represented in the optimal principal components obtained by transforming the subset of features and results are also compared.