Multiple instance classification: Bag noise filtering for negative instance noise cleaning

Data in the real world is far from being perfect. The appearance of noise a common issue that arises limitations data acquisition mechanisms and human knowledge. In classification, label will hinder performance almost all classifiers, inducing bias built model. While has recently attracted researchers’ attention standard it only begun to be studied multiple instance classification. this work, we propose usage filtering algorithms for classification are able reduce impact negative instances within bags. order do so, decompose bags form problem can efficiently treated by specialized filter. Such decomposition tackled different ways, with aim exploiting knowledge offered examples opposite then rebuilt, without identified instances. our experiments, show applying approach diminish even obtain better results at 0% level several classifiers. Our sets out promising dealing datasets further improve rate models.