Algorithms for L1 and Lp Fuzzy c-Means and Their Convergence
Algorithms for L 1 and L p based fuzzy c-means are proposed. These algorithms calculate cluster centers in the general alternating algorithm of the fuzzy c-means. The algorithm for the L 1 space is based on a simple linear search on nodes of step functions derived from derivatives of components of the objective function for the fuzzy c-means, whereas the algorithm for the L p spaces use binary search on the nodes and then the interval to which the cluster center belong. Termination of the algorithms based on different criteria for the convergence is discussed. The algorithm for the L 1 space is proved to be convergent after a finite number of iterations. A numerical example is shown.
KeywordsCluster Center Fuzzy Cluster Binary Search Fuzzy Partition Monotone Nonincreasing
Unable to display preview. Download preview PDF.
- Bezdek, J.C. (1981): Pattern Recognition with Fuzzy Objective Function Algorithms. Plenum.Google Scholar
- Bloomfield, P. and Steiger, W.L. (1983): Least Absolute Deviations: Theory, Applications, and Algorithms, Birkhäuser.Google Scholar
- Bobrowski, L. and Bezdek, J.C. (1991): c-means clustering with the ℓ 1 and ℓ ∞, norms. IEEE Transactions on Systems, Man, and Cybern., 21, 3, 545–554.Google Scholar
- Jajuga, K. (1991): L1-norm based fuzzy clustering. Fuzzy Sets and Systems, 39, 43–50.Google Scholar