Two common data quality issues that can negatively impact the ability of classification models to predict the correct label: class imbalance and class noise. Class imbalance occurs when most instances are belong to a majority class and few instances belong to a minority class. Noise filter is a preprocessing mechanism designed to detect and eliminate noisy instances in the training data. In this paper, we propose a novel approach to eliminate noisy instances from the majority class located inside the borderline area. Our method combine oversampling SMOTE technique with thresholding technique to balance the training data and choose the best boundary between classes. Then a noise detection approach is used to identify and delete the misclassified instances. The proposed method has been examined on several datasets with different imbalance ratio. The results show that our method is robust in identifying noise and improving the AUC scores using random forest as a learning algorithm.