1. The training set all kinds of uneven sample proportion (Unbalanced)
method:
1. oversampling:
Cons: error (noise) samples are likely to have greater impact
2. undersampling:
Cons: throw a larger sample loss
3. To expand the data set:
a. portion of the sample extracted, averaging
b. random noise
(2. If the training set and test set samples themselves a huge gap, the training process are always people suffering)