Crisp Clustering Data in RN C1 C2 Ci Ø Ci Cj = Ø CK Fuzzy Clustering Data in RN C1 C2 Ci Ø Ci Cj Ø CK Spatial Clustering Spatial Clustering Fuzzy C-means The clustering criterion in FCM is Jm(U,V) = S S (uik)m Dik2(vi,xk) U is a fuzzy partition matrix m is the weighting exponent V = [v1,v2,…,vc] is the set of cluster centers (prototype parameters), Dik : distance of xk from center vi Crisp vs. Fuzzy Membership Membership matrix: Uc×n uik is the grade of membership of sample k with respect to prototype i Crisp membership: uik 1, if || xk vi ||2 min || xk v j ||2 j uik 0, otherwise Fuzzy membership: c u i 1 ik 1, k 1,, n Fuzzy c-means (FCM) The objective function of FCM is c n c n m 2 u d u || x v || ik ik k i i 1 k 1 m ik 2 i 1 k 1 FCM (Cnt’d) Introducing the Lagrange multiplier λ with respect to the constraint c u j 1 jk 1, we rewrite the objective function as: c m 2 J uik d ik u jk 1 j 1 i 1 c FCM (Cnt’d) Setting the partial derivatives to zero, we obtain J c u jk 1 0 j 1 J m uikm1 d ik2 0 uik FCM (Cnt’d) From the 2nd equation, we obtain m11 uik 2 m dik From this fact and the 1st equation, we obtain m11 1 u jk 2 j 1 j 1 m d jk C C m 1 m 1 C 1 2 j 1 d jk 1 m 1 FCM (Cnt’d) Therefore, m and 1 m 1 1 1 2 j 1 d jk c uik 2 m dik 1 m 1 1 m 1 FCM (Cnt’d) Together with the 2nd equation, we obtain the updating rule for uik uik 1 1 2 j 1 d jk 1 c d ik2 2 j 1 d jk c 1 m 1 1 m 1 1 m11 2 d ik FCM (Cnt’d) On the other hand, setting the derivative of k with respect to vi to zero, we obtain J 0 vi vi c n m 2 u || x v || ik k i i 1 k 1 u || xk vi ||2 vi k 1 n m ik n uikm k 1 n (xk vi )T ( xk vi ) vi 2 uikm ( xk vi ) k 1 FCM (Cnt’d) It follows that n J uikm ( xk vi ) 0 vi k 1 Finally, we can obtain the update rule of ci: n vi m u ik xk k 1 n m u ik k 1 FCM (Cnt’d) To summarize: n uik 1 d ik j 1 d jk c 2 ( m 1) vi m u ik xk k 1 n m u ik k 1 Fuzzy C-Means Clustering Pros and Cons Advantages Unsupervised Always converges Disadvantages Long computational time Sensitivity to the initial guess (speed, local minima) Sensitivity to noise One expects low (or even no) membership degree for outliers (noisy points) Fuzzy C-Means Clustering Optimal Number of Clusters Performance Index c n min P(c) uikm ( x k vi (c) i 1 k 1 vi x ) n 1 xk Average of all feature vectors x n k 1 c n u i 1 k 1 m ik ( xk vi ) 2 Sum of the within fuzzy cluster fluctuations (small value for optimal c) c 2 2 n u i 1 k 1 m ik ( vi x ) 2 Sum of the between fuzzy cluster fluctuations (big value for optimal c) Fuzzy-Possibililstic C-Means Idea uik is a function of x k and all c centroids tik is a function of x k and v i alone Both are important To classify a data point, cluster centroid has to be closest to the data point Membership For Estimating the centroids Typicality for alleviating the undesirable effect of outliers Fuzzy-Possibililstic C-Means (FPCM), OF and Constraints Objective function c n m 2 min J m, (U, T, V ) (uik tik ) Dik ( U ,T, V ) i 1 k 1 Constraints Membership Typicality c u i 1 ik 1 ,k n t k 1 ik 1 ,i Because of this constraint, typicality of a data point to a cluster, will be normalized with respect to the distance of all n data points from that cluster next slide Fuzzy-Possibililstic C-Means Minimizing OF Membership values Typicality values Same as FCM, but uik resulted values may be different Depends on all data Cluster centers m m vi (uik tik )x k k 1 n tik c Dik j 1 D jk n Dik j 1 Dij 2 m 1 2 1 (u t ) , i k 1 n m ik m ik 1 1 , i, k , i, k Typical in the interval [3,5]
© Copyright 2026 Paperzz