A Comparison of Revision Schemes for Cleaning Labeling Noise

Most pattern recognition systems are built by fitting a model to a training dataset. The accuracy of the resulting classifier is dependent on both the choice of model and the quality of the dataset. However, based on practical experience, most of the standard models (Duda, Hart, & Stork, 2001) tend to have comparable accuracy. In fact, simple models, such as Nearest Neighbor and Decision Trees, have remained very popular among practitioners because they are easy to work with while AbstrAct