作者: Vicente García , Ramón A Mollineda , J Salvador Sánchez , None
DOI: 10.1016/J.KNOSYS.2014.01.021
关键词:
摘要: This paper introduces a framework that allows to mitigate the impact of class imbalance on most scalar performance measures when used evaluate behavior classifiers. Formally, correction function is defined with aim highlighting those classification results present moderately higher prediction rates minority class. Besides, this punishes scenarios are biased towards majority class, but also strongly favor strategy assumes typical task, in which contains relevant samples research purposes. A novel experimental designed show advantages our approach compared standard use well-established measures, demonstrating its consistency and validity.