To identify the relationship between genes and cancers, microarray is always helpful. However, the number of microarray data is quite large, and it is not easy to find out the disease gene from all the microarray data. This paper presents an improved feature selection to filter out the most irrelative or redundant genes. By combining the benefits of "filters" and "wrappers" feature selection, we can not only reduce the processing time of feature selection, but also increase the classification accuracy. In the result, we make a successful result with only 70 genes from 7,129 genes and 70 genes from 12,533 genes in leukemia and Lung cancer microarray data sets. The classification accuracy of leukemia and Lung cancer microarray data are 98.61% and 100%, respectively.
Proceedings of the 2008 International Computer Symposium (ICS 2008)，6頁