Dimensionality Reduction: A Brief Survey on Its Myriad Techniques

Abstract

This paper is a short survey and analysis on different techniques and methods using which efficiency of existing techniques can be improved which are used for data exploration, or search for those features which has deeper relationship amongst the variable, which relies on greater extent on visual methods. However, for multidimensional data set with real time data collected captured using sensors or real time applications, redundant and unused or wanted data can be reduced using dimensionality reduction techniques efficiently as it helps in representing that data more promisingly and representing the reduced data conveniently. It reduces dimensional using features selection and feature extraction. This paper reviews two main techniques used as per its application with limitation for dimensionality reduction they are linear and nonlinear techniques. Linear techniques do linear projection of that data which lies near linear subspace, it results in data compression and increasing data dimensionality. The nonlinear techniques are used as high-dimension data often lies on or much lower dimensional, so, to represent such data low-dimensional data points is used using nonlinear techniques.

Keywords: Dimensionality reduction, principle component analysis (PCA), linear discriminate analysis (LDA), kernel PCA, locally linear embedding (LLE), isomap, local tangent space alignment (LTSE)