The paper investigates some concentration inequalities and excess risk bounds for imbalanced classification problems when using some synthetic oversampling techniques as SMOTE.