An Effective Feature Selection Method Using the Contribution Likelihood Ratio of Attributes for Classification

Authors:
Zhiwang Zhang;Yong Shi;Guangxia Gao;Yaohui Chai
Affiliations:
School of Information of Graduate University of Chinese Academy of Sciences, China Chinese Academy of Sciences Research Center on Fictitious Economy and Data Science, Beijing, China (100080);Research Center on Fictitious Economy and Data Science, Chinese Academy of Sciences, Beijing 100080, China/ College of Information Science and Technology, University of Nebraska at Omaha, Omaha, U ...;Foreign Language Department, Shandong Institute of Business and Technology, Yantai, China 264005;School of Management of Graduate University of Chinese Academy of Sciences, China Chinese Academy of Sciences Research Center on Fictitious Economy and Data Science, Beijing, China (100080)
Venue:
Advanced Web and NetworkTechnologies, and Applications
Year:
2008

Citing 4
Cited 0

Machine Learning

Machine Learning
Data Mining: Concepts and Techniques

Data Mining: Concepts and Techniques
Data Mining and Knowledge Discovery Handbook

Data Mining and Knowledge Discovery Handbook
Data Mining: Practical Machine Learning Tools and Techniques, Second Edition (Morgan Kaufmann Series in Data Management Systems)

Data Mining: Practical Machine Learning Tools and Techniques, Second Edition (Morgan Kaufmann Series in Data Management Systems)

Quantified Score

Hi-index	0.00

Visualization

Abstract

Feature selection is a very crucial step in data mining process. It aims to find the most important feature subset from a given feature set without degradation of classifying information. As for the traditional feature selection method, the number of candidate feature subsets created by algorithm in an iterative computational way is exponential in the size of the initial attribute set. And relevant algorithm occupies a lot of the system resources in time and space. In this paper, we study and develop a novel feature selection method and provide its mathematic principle, which is based on the factors of attributes contributing to target attribute and their maximum information divergence value (MIDV) to select small enough feature subset and improve the classification accuracy. And then the extensive experiment shows that our proposed method is very efficient in computational performance and scalability than traditional methods.