Feature weighting for Centroid Neural Network

A feature weighting procedure for centroid neural network (FWP-CNN) is proposed in this paper. The proposed FWP-CNN evaluates the importance of each feature in data by introducing a feature weighting concept to the CNN in the proposed algorithm. The use of feature weighting makes it possible to reject noises in data and thereby achieves a better clustering performance. Experimental results on a synthetic data set show that the proposed FWP-CNN outperforms conventional algorithms including the k-means algorithm, self-organizing map(SOM), and CNN in terms of the clustering accuracy.