how can do training with libsvm in a right way?

hi everybody I want to use libsvm. for training I used:

    model = svmtrain(YTrain, XTrain, 'libsvm_options')

I have about 10000 samples. when I used libsvm with this samples I see this result:

I reduced sampels to 100 I saw this result:

can you help me where is the problem? are the features that I selected good? should I change C and gamma? how can I change them? Is this overfitting? I appreciate your helps and supports in advanced.