Improved Self-Generating Prototypes Algorithm for Imbalanced Datasets

摘要: Some real world datasets have different proportions of classes, too many instances the majority classes and only a few minority those are called imbalanced datasets. Many applications, like medical diagnosis risk analysis, interested in under-represented class, but classifiers prototype generation techniques usually bias towards classes. Because that, problem classification with has become an important topic Pattern Recognition. The Self-Generating Prototypes (SGP) high reduction power excelent performance balanced datasets, but, generated prototypes do not good representation training dataset. This algorithm generates few, or even none, aim this paper is to propose Adaptive (ASGP), improvement SGP2, second version SGP, designed handle also exposes reasons for low SGP2 such Empirical results show that ASGP higher than especially when it comes accuracy