The Comparison of Different Feature Decreasing Methods Base on Rough Sets and Principal Component Analysis for Extraction of Valuable Features and Data Classifying Accuracy Increasing

The primary purpose of the data mining is extraction of required information from a huge amount of datasets. In this regard, it must be tried to omit invalid, noisy and incomplete information as far as possible. Our results and assessment from data mining process would be incomplete, while this info...

Full description

Saved in:
Bibliographic Details
Main Authors: Lotfabadi, Maryam Shahabi, Moghadam, Amir Masoud Eftekhari
Format: Conference Proceeding
Language:English
Subjects:
Accuracy
Approximation methods
Artificial neural networks
Feature extraction
Fuzzy sets
Principal component analysis
Rough sets
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The primary purpose of the data mining is extraction of required information from a huge amount of datasets. In this regard, it must be tried to omit invalid, noisy and incomplete information as far as possible. Our results and assessment from data mining process would be incomplete, while this information is not omitted totally. It means, if the creditable value according required content has not been defined for the information, there is the uncertainty problem for extracted data. One of the ways for decreasing the redundant and invalid features of data is rough sets, rough fuzzy sets and the principal component analysis methods. In this paper, these decreasing methods versus some other decreasing methods on UCI datasets have been compared. All the mentioned decreasing algorithms have been run and used on 15 training data sets of UCI machine. The classification accuracy of most data sets has not been assessed yet by these decreasing methods. The two classifiers of the support vector-machine with RBF kernel and the neural network have been used. Base on the tests, the decreasing method by rough fuzzy set and support vector-machine classifier provide better results, i.e., a classifying accuracy of about 94.87%.
DOI:10.1109/ICIIC.2010.11