28 An Application of a New Hybrid for Feature Selection Using Colorectal Cancer Microarray Data
-
Published:2008
Download citation file:
The objective of this work is to implement a new hybrid feature selection system comprised of a genetic algorithm (GA) and a support vector machine program termed SVMperf. We have used this system to perform feature reduction of a colorectal cancer microarray dataset generated by the Moffitt Cancer Center. Using variance pruning as a coarse feature selection process with the GA-SVMperf wrapper, the method provided an Az (performance measure) value of .97 with only 7 features after 30 generations. Using a combination of variance pruning, t-tests and the GA-SVMperf wrapper, the method provided an Az value of 1 with only 6 features. These results show tremendous improvement over other linear techniques that were used to do feature reduction. This implies that the technique used in this paper is best suited for feature reduction since it can find patterns that separate different classification cases with high success.