Improved Similarity Based Fuzzy C-Means Clustering Algorithm

To solve the shortcoming of the fuzzy c-means clustering algorithm(FCM) that fails to cluster the non-convex datasets effectively, we propose an improved fuzzy c-means clustering algorithm(IFCM). IFCM first constructs the nearest neighbor network graph by finding the r neighbors of each data point through the 3N algorithm, if a connected relationship exists on two data points in the graph, the Euclidean Distance is employed to measure the distance between them; Otherwise, it performs a more reasonable similarity calculation strategy, which merges the distance value mentioned above with the Euclidean Distance between the other two data points which have the furthest distance in the dataset. Meanwhile, it updates the membership degree of each data point in each cluster and the center of each cluster. IFCM will repeat the process mentioned above until all the data points in the dataset have been classified properly. Further analysis on artificial and UCI datasets verifies the importance of similarity strategy improvement for clustering, justifying the rationality and effectiveness of IFCM.