How can I improve my code?

%Step1:Load a dataset
load thyroid_dataset;
inputs = thyroidInputs;
targets = thyroidTargets;

%Step2:create Multi ANN
hiddenLayerSize = 20;
Numb_Ag = 7; %num of agents
Numb_round = 2;

for i=1:Numb_round
   for j=1:Numb_Ag
    %create a pattern recognition network
     net= feedforwardnet(hiddenLayerSize);
     %Step3:Training
    [Anet{i},tr{i}] = train(net,inputs, targets); %Training
    vperf(i,j)= (1-tr{i}.best_vperf)*100;
        if (j==Numb_round) && (vperf(i,j)< vperf(i,j))
          while(1)
              net= feedforwardnet(hiddenLayerSize);
              [Anet{i},tr{i}] = train(net,inputs, targets); %Training
              vperf(i,j)= (1-tr{i}.best_vperf)*100;
                if vperf(i,j)>= vperf(i,1)
                  break;
                end
           end
        end
   end
end
[verf BestAg]= max(vperf(:,2));
net = [];
net = Anet{BestAg,1}
output = net(inputs);

%steps4 : zero -sum strategy

     [verfBestANN_id] = max(vperf);

%Steps 5 Test with best ANN
   outputs = net{BestANN_id}(inputs);
   accuracy_performance = (1-perform(net{BestANN_id}, targets, outputs))*100 ;

How can I improving this code in terms of algorithm design and analysis?