Mis-classified instance learning and recovery in classification
Yun Zhu, Yanqing Zhang, Yi Feng Pan · 2012
Many classification models have been proposed in past few decades. Lots of variations based on those models are also developed for better performance. Instead of model tuning or modification, we achieve higher classification accuracy by analyzing the dataset and recovering the instances that are mis-classified by the given classifier. We develop three metrics to identify those mis-classified instances. Experiments show our method can obtain performance improvement with the chosen classifier in multiple datasets.