1. INTRODUCTION
When data from multi-dimensional measurements is represented as a feature vector, the feature space of the raw data often has a very large dimension. It is usually prudent to re-represent the original measurements in more compact form (a shorter feature vector). The process of selecting features from the raw measurements is one of feature selection or feature extraction. It is, in general, a problem of dimensionality reduction. The ultimate goal of feature selection/extraction is to find the minimum number of features required to capture the essential structure in the raw data. This minimum number of features is termed the intrinsic dimensionality of the data. This dimensionality reduction is accomplished by applying a transformation (linear or non-linear) to the the input data.