The class imbalance problem emerged strong as it extended more into real domains. A dataset is said to be imbalanced if the classification categories are not approximately equally represented. Accuracy, which is considered as the major performance measure of classifier is not appropriate for imbalanced datasets as the cost of errors vary markedly. This paper proposes an efficient oversampling technique based on SMOTE (Synthetic Minority Oversampling Technique) which is used for generating synthetic samples in the process of balancing the dataset along with an editing technique for efficient feature extraction based on Rough Set Theory. Performance measures which are more appropriate for imbalanced datasets such as ROC curves and cost curves are considered along with accuracy.