Fuzzy clustering of fuzzy data considering the shape of the membership functions using a novel representation learning technique

Most existing distance measures for fuzzy data do not capture differences in the shapes of the left and right tails of membership functions.As a result, they may calculate a distance of zero between fuzzy data even when these differences exist.Additionally, some distance measures cannot compute distances between fuzzy data when their membership functions differ in type.In this paper, inspired by human visual perception, we propose a fuzzy clustering method for fuzzy data using a novel representation technique that is capable of detecting small differences in the shapes of the left and right tails of membership functions.Moreover, it effectively clusters fuzzy data even when their membership functions differ in type.By utilizing the pretrained ResNet50 network as a feature extractor and applying the FCM clustering method to the output from the last convolutional layer, our approach achieves high accuracy in clustering both synthetic and real data sets.Experimental results demonstrate that our method achieved a Rand Index of 0.9965, outperforming state-of-the-art methods, making it particularly suitable for applications that require high clustering accuracy.