Enhancement of PCA-Based Dimensionality Reduction Using BB-BC Optimization Algorithm

Abstract

Dimensionality reduction is an important step for various applications where usage of the internet and multimedia systems is involved which requires huge bandwidth and storage space. This paper presents Big Bang-Big Crunch (BB-BC) optimization algorithm based two new approaches to feature selection for dimensionality reduction. In first approach, PCA is used to extract the features (eigenvectors) and defines feature set based on computation of knee point and BB-BC optimization algorithm, in turn selects an optimal subset from the predefined feature set. In the second approach, PCA is used only for feature extraction and BB-BC optimization algorithm is used for optimal feature selection from all extracted features. Olivetti Research Laboratory (ORL) face database has been used for performing the experiment and recognition rate is the parameter to be optimized for face recognition as an application area. The experimentation proves BB-BC as a powerful soft computing technique to solve such NP hard problems.