This paper presents a data classification problem and methods toimprove imbalanced data classification. Especially, biomedical data has a very highimbalance rate and the sample identification of minority class is a very important.Many studies have shown that border elements are important in imbalanced dataclassification such as Borderline-SMOTE, Random Under Border Sampling. Thispaper provides a new method of adjusting data generating synthetic elements onthe borderline of the minority class, identify and eliminate noise elements of themajority class to achieve better classification efficiency. Experimental results ofclassification of SVM algorithm on six datasets of UCI international standard datawarehouse Blood, Haberman, Pima, Yeast, Ionosphere, and Glass showed that theadjustment of borderline has a positive effect on classification and the results areconsidered statistically significant.