Enhancing Confusion Entropy as Measure for Evaluating Classifiers

Performance measures are used in Machine Learning to assess the behaviour of classifiers. Many measures have been defined on the literature. In this work we focus on Confusion Entropy (CEN), a measure based in Shannon's Entropy. We introduce a modification of this measure that overcomes its disadvantages in the binary case that disables it as a suitable measure to compare classifiers. We compare this modification with CEN and other measures, presenting analytical results in some particularly interesting cases, as well as some heuristic computational experimentation.