how to update weight in neural network using Error back propagation algorithm?

Good Morning, I solve the problem using error back-propagation algorithm for classification of imbalanced data. In my work i have do Ann-thyroid data is transformed into two -class problems."Ann-thyroid13(23)" refers to a class1(2) is the minority class while class 3is treated as the majority class.My question is how to update weight in my problem. x=textread('D:\UCI\thyroid-disease\ann-train.data');%read whole document % retrieve the last coloum value lst=x(:,22); [m n]=size(x); k=1;%store Minority class1 for i=1 : m if (x(i,22)==1) a(k,:)=x(i,:); k=k+1; end end [m1 n1]=size(a);

     k1=1;%store Majority class3
      for i=1:m
         if(x(i,22)==3)
            a1(k1,:)=x(i,:);
            k1=k1+1;
         end
     end
     [m2 n1]=size(a1);
    k2=1;%store Minority class2
   for i=1:m
      if(x(i,22)==2)
        a2(k2,:)=x(i,:);
        k2=k2+1;
      end
   end
   [m3 n1]=size(a2);
    ann13=vertcat(a,a1);
    ann23=vertcat(a2,a1);
    [m4 n1]=size(ann13);
disp('Number of Input Nodes 21');
disp('Number of Hidden Nodes 16');
 disp('Number of Output Nodes 2');
 n=21;
 h=16;
 m=2;
 Tp=3581;%Training pattern
 disp('Generate weights between -.0001 to .0001 Randomly');
  w=linspace(-.0001,.0001,336);
  w1=reshape(w,h,n);
  disp('Input layer to Hidden Layer weights');
  disp(w1);
  for i=1:n
    for j=1:h
        h1=tanh(w1(j,i)*ann13(j,i)/2);
     end
  end
 disp('Hidden Layer values');
disp(h1);
  v=linspace(-.0001,.0001,32);
  v1=reshape(v,m,h);
  disp('Hidden Layer to Output Layer weights');
  disp(v1);
  for j=1:h
    for k=1:m
        y(k)=tanh(v1(k,j)*h1/2);
     end
 end
 disp('Output Layer values');
disp(y(k));
[yr yc]=size(y);
for i=1:n1
     for j=i:n1
if ann13(j,i)==t1(j,i)
    tk=1;
else 
    tk=-1;
end
    end
 end
for p=1:Tp
    for k=1:m
        if tk==1
             e=1/2*sqrt(t1(k)-y(k));
        else
             e=1/2*sqrt(t(k)-y(k));
        end
    end
end
disp('The conventional error function of p pattern');
 disp(e);

 neta=0.001*[(n+1)+(m+1)]/2;%Learning Parameter
 for i=1:n
     for j=1:n
         if tk==1
             w1=neta(t1(j,i)-h1(j,i)*y(j,i)*ann13(j,i));
          else
              w1=neta(t1(j)-h1(j)*y(j)*ann13(j,i));
          end
     end
 end   
     disp(w1);