Dear Vahid

Try to use K fold cross validation partition;

or try to split your dataset into three groups: training, validation, testing with 0.60, 0.20, 0.20 respectively.

both of these method are a good to avoid bias.

Regards

Hello

Such skewed data may causes the accuracy paradox problem and bias against some classes when evaluating the performance. I recommend Synthetic Minority Over Sampling Technique (SMOTE), Please read pages 3-4 here:

Article Optimizing Stochastic Gradient Descent in Text Classificatio...

Dear Vahid

Try to use K fold cross validation partition;

or try to split your dataset into three groups: training, validation, testing with 0.60, 0.20, 0.20 respectively.

both of these method are a good to avoid bias.

Regards

@fallah

Thanks for your answering.Use by your solution there will be the imbalance problem still.

Please, see article (attached) :

Jason Van Hulse, Taghi M. Khoshgoftaar, Amri Napolitano "Experimental Perspectives on Learning from Imbalanced Data" Proceedings of the 24th International Conference on Machine Learning, Corvallis, OR, 2007

There are many techniques to deal the problem of imbalance data:

1. SMOTE

2. Up-sampling

3. Down-sampling etc.

I'll personally suggest to use another technique GSS (Gaussian scale space theory, along with data augmentation techniques on class-1 to augment the dataset upto double to make it [1:1] ratio.