How are you performing your experiments?  Neural networks are often initialized randomly so this will likely create different results for different training runs.  If you are changing the training or test sets in anyway (such as by doing cross-validation), then this will cause differences in the resulting confusion matrix as well.

Thank you for reply Sir,

I am performing my experiments by matlab commands.

the commands are

 load diabetes;

 input = diabetes (1:8, :);

 target = diabetes (9, :);

 hidden= 40;

 net = newcf (input, target, hidden );

 net.divideParam.trainratio = 50/100;

 net.divideParam.testratio = 50/100;

 [net,tr] = train(net,input,target);

output = sim(net,input);

 plotconfusion(target,output)

the classification result is different everytime.

Did you try to fix the seed?

It works

Thank you Sir.