Improved SMOTE method and system based on adaptive neighborhood size

Abstract

The invention relates to an improved SMOTE method and system based on adaptive neighborhood size, and the method comprises the steps: employing a different neighborhood value for each piece of minority of data, and automatically determining the neighborhood value through tracking a precision reduction point in a super-rectangular region formed by a minority samples and an adjacent sample point. The synthesized data can be placed in the rectangular region, random linear interpolation is not needed, and as a minority class in the formed super-rectangular area is dominant, the synthesized samplegenerated in the rectangular region is safer and more reasonable.

Description

technical field [0001] The invention relates to the technical field of machine learning and data mining, in particular to an improved SMOTE method and system based on adaptive neighborhood size. Background technique [0002] With the rapid development of information technology, the amount of data accumulated in various industries is growing explosively, and the problem of class imbalance has attracted widespread attention. The class imbalance problem exists in many application fields, such as biomedical diagnosis, multimedia data classification, insurance fraud detection, spam and web page identification, credit card and telecom fraud detection, etc. In the class imbalance problem, we pay more attention to the minority class Data recognition rate. [0003] At present, class imbalance learning methods can be mainly divided into three levels: data, features, and algorithms. The main method at the data level is the sampling method based on data distribution. In recent years, ...

AI Technical Summary

Problems solved by technology

The same problem occurs when the distribution of the minority class is irregular and complex, figure 1 Middle 3 is a complex situation. Since the positive class forms a non-convex cluster, other methods such as SMOTE to synthesize positive class samples will generate synthetic samples in the negative class area, and these synthetic samples will increase false positives.

Images