Fast Correlation-Based Filter Software for Feature Selection

Arizona State University, Computer Science and Engineering, Data Mining and Machine Learning

Overview

Feature selection is a preprocessing technique frequently used in data mining and machine learning tasks. It can reduce dimensionality, remove irrelevant data, increase learning accuracy, and improve results comprehensibility. FCBF is a fast correlation-based filter algorithm designed for high-dimensional data and has been shown effective in removing both irrelevant features and redundant features.

About FCBF Software

This software package is prepared in Java. It is provided free of charge to the research community as an academic software package with no commitment in terms of support or maintenance. Data may need to be discretized before running FCBF. Please refer to the paper below. A discretization package is also available on our web site.

Downloads

Java package can be downloaded here.
Before 3/15/04 a version for Linux was released.

        The software is freely available for academic use, and can be downloaded here for Linux
        Run "tar xvfz FCBF.tar.gz" to obtain all the files including README and some sample data files.
        README provides the details about how to run, how to prepare the data, and how to read the results.

People

ASU Data Mining and Machine Learning Lab: Huan Liu, Lei Yu
Ravi Bhimavarapu, Manoranjan Dash, Farhad Hussain

References

L. Yu and H. Liu. "Feature Selection for High-Dimensional Data: A Fast Correlation-Based Filter Solution". In Proceedings of The Twentieth International Conference on Machine Leaning (ICML-03), pp 856-863, Washington, D.C., August 21-24, 2003. pdf

H. Liu, F. Hussain, C.L. Tan, and M. Dash. "Discretization: An Enabling Technique", Journal of Data Mining and Knowledge Discovery 6(4): 393-423; Oct 2002.

Created on 3/15/2004

Updated on 4/14/2004