Feature weighting based on L-GEM
Qiancheng Wang, Wing W. Y. Ng, Patrick P. K. Chan, Daniel Yeung · 2010
In this paper, we propose a novel method to weight features for their relevance to the given classification problem. The weight of a feature is computed by its Localized Generalization Error model (L-GEM). Then, a Radial Basis Function Neural Network (RBFNN) is trained by those weighted features. Experimental results on image classification problem show that the proposed method is efficient and effective in comparison to current methods.