How can I improve my prediction accuracy (93%) using NEWRBE function because my target consists of 0's or 1's?

Hi, I am trying to get 100 accuracy for predicted data. However, I am getting 93.75. I am using NEWRBE function because my targets are 0's or 1's in the output. I would highly appreciate, if anyone could guide me on this problem.

details about my database:

[M N] = size(in); %[18 258]
[I N] = size(out); %[1 258]
net = newrbe(in,out,1e-05);
y   = sim(net,in)';
100-100*sum(abs(y'-out))/length(out)

ANS = 100. %accuracy is 100 when i am not using prediction.

When i divide the data set that is 75% is trained and 25% is untrained and used as prediction. following is the result.

[m n] = size(in1) % trained data, m =18 and n = 194
newrbe(in1,out1,1e-05)
y   = sim(net,in1)';
100-100*sum(abs(y'-out1))/length(out1)

ANS = 100. (accuracy is still 100).

However, when i predict rest of the data I get accuracy of 93.

[m n] = size(in2); %untrained data m = 18 and n = 64
y   = sim(net,in2)';
100-100*sum(abs(y'-out2))/length(out2)
ANS = 93.75

When i check figures. in trained network, all 0's and 1's are matched perfectly to outputs. But in untrained network 0's are still 1's (error). HOW TO FIX, SUCH THAT I GET ACCURATE RESULT USING NEWRBE OR NEWRB??

figure(1) shows trained network and figure(2) shows predicted network result (untrained).