Effective gene selection method with small sample sets using gradient-based and point injection techniques

D. Huang, T. W. S. Chow

Two properties of our gene selection scheme

Gradient-based strategy

It originates form the optimization theory, and enable gene searching to be conducted along the possible steepest optimization direction.

Point-injection strategy

It is motivated by the noise-injection techniques related to classification learning and the smooth error evaluation scheme.

The source code (Matlab file) about our methods is zipped as demo.zip.

More experimental results:

Blow, SFS -- the conventional sequential forward searching scheme;

SFS-G -- SFS + gradient-based strategy;

WMSFS -- SFS + two strategies, which is defined in the submitted paper.

Clearly, the differences between SFS and SFS-G demonstrate the impact of the gradient based strategy, while the differences of WMSFS from SFS-G indicate the contribution of the point-injection strategy.

1. The comparisons results on the synthetic data mentioned in the submitted paper.

2. The comparisons results on colon cancer classification data.

3. The comparisons results on prostate cancer classification data.