An instance-oriented performance measure for classification

Shuang Yu, Xiongfei Li, Yuncong Feng, Xiaoli Zhang, Shiping Chen

Abstract: Highlights•Propose a classifier evaluation metric from the perspective of instances.•Propose a selection criteria for acceptable classifiers based on the new metric.•Propose an approximate strategy to improve the computation efficiency of difficulty.•Evaluations on the new metric and acceptable classifiers confirm their effectiveness.•Evaluation on the approximate difficulty confirms its feasibility.