Collaborative Optimization of Clustering by Fuzzy c-means and Weight Determination by ReliefF

The ReliefF algorithm is an important attribute weighting approach, which is built on the basis of classification labels. And the attribute weights of weighted FCM (WFCM), a popular fuzzy clustering algorithm, can be gotten by ReliefF. In the light of the idea of collaborative learning, a collaborative optimization of clustering by fuzzy c-means and weight determination by ReliefF (Co-WFCM) is introduced in this paper, in which FCM/WFCM and ReliefF who act as unsupervised and supervised learners are trained reciprocally. Experimental results show that the algorithm is helpful to get more satisfying clustering results and more rational attribute weights in some cases. And on the other hand, some suggestions for applicability of the ReliefF+FCM/WFCM algorithm framework can be given by analysis of the attribute weight sequences.