Automation and Computational Intelligence for Road Maintenance and Management

Appendix CSample Code for Feature Selection

function [data_mod, index_rem,SIM,SIM1,H]=feat_sel_sim_Schweizer_1(data, p)

%    p in (0, \infty) as default p=1. if nargin<3
p=1;
end

l=max(data(:,end)); % #-classes m=size(data,1); % #-samples t=size(data,2)-1; % #-features dataold=data;
tmp=[];
% forming idealvec using arithmetic mean idealvec=zeros(l,t);
for k=1:l idealvec_s(k,:)=mean(data(find(data(:,end)==k),1:t));
end

%scaling data between [0,1] data_v=data(:,1:t); data_c=data(:,t+1); %labels mins_v = min(data_v);
Ones = ones(size(data_v));
data_v = data_v+Ones*diag(abs(mins_v)); for k=1:l
tmp=[tmp;abs(mins_v)]; end
tmp;
idealvec_s = idealvec_s+tmp; maxs_v = max(data_v);
data_v = data_v*diag(maxs_v.^(-1)); idealvec_s=idealvec_s./repmat(maxs_v,l,1); data = [data_v, data_c];
% sample data datalearn_s=data(:,1:t); land=0.1;
% similarities sim=zeros(t,m,l); for j=1:m

for i=1:t
for k=1:l
%%similarity Luca    with p=1
sn_1(i,j,k)=1-(max(0,((1-(1-datalearn_s(j,i)))^p+((1-idealvec_s(k,i))^p)-1)))^(1/p);
sn_2(i,j,k)=1-(max(0,((1-(datalearn_s(j,i)))^p+((1-(1-idealvec_s(k,i)))^p)-1)))^(1/p);
sim(i,j,k)=(max(0,((sn_1(i,j,k))^p+(sn_2(i,j,k))^p))-1)^(1/p); end
end end
%class 1,2,3,4,. n
% SIM(:,:,1) = similarity for class 1
% SIM(:,:,2) = similarity for class 2 SIM=sim;
% reduce number of dimensions in sim sim=reshape(sim,t,m*l)';
SIM1=sim;

% possibility for two different entropy measures
% moodifying zero and one values of the similarity values to work with
% De Luca's entropy measure