Research Of Fuzzy Clustering Based On Fuzzy Quotient Space

When dealing with complex problems, fuzzy quotient space granular computing has a strong advantage over quotient space model. Under the hierarchical structure of fuzzy quotient space, problems can be analyzed in different levels. So how to choose a proper level to slove problems is an important research direction in fuzzy granular computing.In terms of choosing the best level in a hierarchical structure of fuzzy quotient space, a granularity-based validity index is proposed, which has fully considered the relationship among the data items with the normalized metric distance. The granularity-based validity index overcomes the shortcoming of the traditional indexes. By using it, an optimal level is determined as the final clustering results.The traditional fuzzy c-means(FCM) algorithm is sensitive to initial clustering centers and needs to specify the number of clusters in advance. When dealing with the unbalanced classified samples, it is also difficult to obtain the correct clustering results. By introducing fuzzy similarity relation and normalized metric distance, a hierarchical structure of fuzzy quotient space is constructed. And then, through the granularity-based validity index, it can choose the best level to determine the number of clusters, and select samples with high similarity as the initial cluster centers. Finally, an improved fuzzy quotient space of FCM(IFQFCM) is developed. In comparison with traditional algorithms, the improved algorithm has fewer numbers of iterations and higher accuracy. At the same time, this method is more suitable for problems with unbalanced classified samples. Experiment shows the effectiveness of the IFQFCM algorithm.