Man ,data already balanced befor modilng ,do you mean that data balncing sholud be done inside model with cross val for each model will be used ?

Judging from what you said, that means you used SMOTE to balance the classes, SMOTE OVERSAMPLING should be deone before training the model and only on the train dataset

Yes man smote don on training data set only but result not good as i metioned before ,i thing bbad result due to train data need to increased , is this bad reasult isnormal and more than one classes predicted

Try changing your evaluation metric

Thanks man