Currently, financial fraud is increasingly common and cause serious consequences. Therefore, detecting and preventing financial fraud has attracted great attention from researchers. Today, the problem of detecting financial fraud can be resolved with the support of the data mining techniques. In it, the subclass is a supervised learning method is to apply the most common. However, due to the particularities of financial data, the number of samples to be determined is a lot less fraud than the valid samples, leading to imbalanced condition data. Some popular methods to solve this problem as SMOTE, borderline-SMOTE has achieved positive results. In this paper, we propose a new method, MASK, in order to change the label of the part from the majority of the class based on the distribution of the density of the element class minority. The experimental results also indicated a new and effective method of improving accuracy of detection
đang được dịch, vui lòng đợi..
