1function [DBI]=DavisBouldinIndex(d,c,kk); 2% Davis-Bouldin-Index is a Cluster separation index (CSI) 3 4% KAPPA.M estimates Cohen's kappa coefficient 5% 6% [kap,sd,H,z,OA,SA,MI] = kappa(d1,d2); 7% [kap,sd,H,z,OA,SA,MI] = kappa(H); 8% 9% d1 data of scorer 1 10% d2 data of scorer 2 11% 12% kap Cohen's kappa coefficient point 13% se standard error of the kappa estimate 14% H data scheme (Concordance matrix or confusion matrix) 15% z z-score 16% OA overall agreement 17% SA specific agreement 18% MI Mutual information or transfer information (in [bits]) 19% 20% Reference(s): 21% [1] Bezdek, J.C.; Pal, N.R.; Some new indexes of cluster validity 22% Systems, Man and Cybernetics, Part B, IEEE Transactions on Volume 28, Issue 3, June 1998 Page(s):301 - 315 23% 24% 25 26% http://www.ucl.ac.uk/oncology/MicroCore/HTML_resource/distances_popup.htm 27 28% $Id$ 29% Copyright (c) 2006 by Alois Schloegl <alois.schloegl@gmail.com> 30% This is part of the BIOSIG-toolbox http://biosig.sf.net/ 31 32% This library is free software; you can redistribute it and/or 33% modify it under the terms of the GNU Library General Public 34% License as published by the Free Software Foundation; either 35% version 2 of the License, or (at your option) any later version. 36% 37% This library is distributed in the hope that it will be useful, 38% but WITHOUT ANY WARRANTY; without even the implied warranty of 39% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 40% Library General Public License for more details. 41% 42% You should have received a copy of the GNU Library General Public 43% License along with this library; if not, write to the 44% Free Software Foundation, Inc., 59 Temple Place - Suite 330, 45% Boston, MA 02111-1307, USA. 46% 47 48 49CL = unique(c); 50M = length(CL); 51 52t = 2; q = 2; 53D = zeros(M,M); 54 55for k=1:M, 56 %v(k,:)=mean(dk,1); % center of each cluster 57 [dk, v(k,:)] = center(d(c==CL(k),:),1); 58 S(k) = sum(sqrt(sum(dk.^2,2)).^q).^(1/q); 59end; 60 61for k1=1:M, 62for k2=1:M, 63 d(k1,k2)= sum(abs(v(k1,:)-v(k2,:)).^t).^(1/t); % Minkowski Distance of order t 64end; 65end; 66 67[x,y] = meshgrid(S); 68DBI = mean(max((x+y)./D + diag(repmat(NaN,1,M)))); 69 70 71