Further improvements in Feature-Weighted Fuzzy C-Means

In cluster analysis, certain features of a given data set may exhibit higher relevance than others. To address this issue, Feature-Weighted Fuzzy C-Means (FWFCM) approaches have emerged in recent years. However, there are certain deficiencies in the existing FWFCMs, e.g., the elements in a feature-weight vector cannot be adaptively adjusted during the training phase, and the update formulas of a feature-weight vector cannot be derived analytically. In this study, an Improved FWFCM (IFWFCM) is proposed to overcome these shortcomings. The IFWFCM_KD based on the kernelized distance is also proposed. Experimental results reported for five numerical data sets and the color images show that IFWFCM is superior to the existing FWFCMs. An interesting conclusion, that IFWFCM_KD might not improve the performance of IFWFCM, is also obtained by applying IFWFCM_KD to tackle the above-mentioned numerical data sets and color images.