Improved SMOTE algorithm for imbalanced dataset

When applying traditional classifiers to imbalanced dataset, the result might be bias towards the majority class, which leads to poor performance of classifiers. Synthetic Minority Oversampling Technique(SMOTE) is a popular algorithm to improve the classifier's performance through generating new minority samples and making dataset balanced. Based on SMOTE, two new over-sampling algorithms DSMOTE and ESMOTE are proposed in this paper. Being different with SMOTE which treats all minority samples equally, the two new over-sampling algorithms mainly synthesize new samples near the easily misclassified samples to improve the classification accuracy of minority class. Experiments show that DSMOTE and ESMOTE could both get better performance than SMOTE.