Optimal Feature Subset Selection Using Differential Evolution and Extreme Learning Machine
Journal: International Journal of Science and Research (IJSR) (Vol.3, No. 7)Publication Date: 2014-07-05
Authors : Bharathi P T; P. Subashini;
Page : 1898-1905
Keywords : Feature Selection; Gray Level Co-occurrence Matrix; Differential Evolution Feature Selection; Extreme Learning Machine;
Abstract
Feature selection problem often occurs in pattern recognition and more specifically in classification. Features extracted from feature extraction methods could contain a large number of feature set. In original feature set, some of them can prove to be irrelevant, redundant and even unfavorable to classification accuracy. So it is essential to remove these type of features, which in turn leads to dimensionality reduction and could eventually improve the classification accuracy. The objective of feature selection is carried out in three steps. Firstly, improving the prediction performance of the predictors, secondly for providing faster and more cost-effective predictors, and finally providing a better understanding of the underlying process that generated the data. Considering river ice images, analyzing the different types of ice features and their characteristics is complex in nature. Hence, in this work feature extraction is carried out by computing Gray Level Co-occurrence Matrix (GLCM) features for 00, 450, 900 and 1350 and feature subset selection is performed with Differential Evolution Feature Selection (DEFS) algorithm. DEFS, utilizes the DE float number optimizer in a combinatorial optimization problem like feature selection. DEFS feature selection method highly reduces the computational cost while at the same time proves to present a powerful performance and provides 93 % accuracy with the features selected. Proposed method, Extreme Learning Machine (ELM) combined with DEFS (ELM_DEFS) technique selects best feature subset from the original feature set. Selected feature set is used for better simplification and training the classifier, to classify river ice types correctly. Features selected from the proposed method reduce 65 % of the features and provide 97.78 % accuracy for river ice images.
Other Latest Articles
- Detection of Node Replication Attack in Wireless Mobile Sensor Network Using Paillier
- Design and Development of a Novel Technique to Reduce Inter-Symbol Interference Using Decision Feedback Equalization
- Morphological Structure of Tegument in Fasciolahepatica Affecting Sheep in Saudi Arabia
- Comparative Analysis of Various Image Denoising Techniques: A Review Paper
- A Novel Technique for the Estimation of Accurate Thickness of the Sea Ice
Last modified: 2021-06-30 21:02:23