Fuzzy clustering using extended MFA for continuous-valued state space
Min-Hee Kim, Heesook Choi, Keon Myung Lee · 2003
In classical clustering, an item must belong to any one cluster, whereas fuzzy clustering describes more accurately the ambiguous type of structure in data. MFA (mean field annealing) combines characteristics of simulated annealing and a neural network, and exhibits the rapid convergence of the neural network, while preserving the solution quality afforded by SSA (stochastic simulated annealing). An extended MFA algorithm to solve the fuzzy clustering problem is proposed. It has continuous-value state space. The results of the experiment are given and compared with those of the fuzzy ISODATA algorithm. Fuzzy clustering using the MFA algorithm shows a lower energy state than that of the fuzzy ISODATA algorithm. The perturbing of only one variable is simpler and faster than traditional SSA method to perturb all the variables together, and ultimately enables true parallelism.>