Feature grouping technique to relax sample support requirement for sparse linear feature extraction

In this paper, the employment of feature grouping is concentrated on to relax sample size requirement of sparse linear feature extraction. Genetic Linear Projections (GLP) is one of the latest linear feature extraction approaches. The major drawback of applying GLP to sparse linear feature extraction is that it may encounter a small sample size problem. In this work, GLP is reviewed in a practical, problem-oriented context, and then goes to examine potential application in sparse linear feature extraction to abate the required sample support by combining the feature grouping techniques with the evolutionary optimizer based linear feature extraction approach. Simulation results manifest that the satisfying performance of sparse linear feature extraction can be obtained by using an alternative approach to try to describe the sparse structure in the scenario where the sample size is small.