What's wrong with hit ratio?

The views of author on issues related to faults in hit ratio and its magnitude are presented. He has evaluated the magnitude through an empirical experiment on three multivalued classification data sets, using two machine learning models. The concept of hit ratio is problematic as it doesn't compensate for classifications caused by chance. ROC curves offer an alternative to using hit ratio are useful visualization tools for analyzing trade-offs between true positives and fast positives. A classifier that does better than randon yields a concave ROC curve above that straight line. By computing the area between these two lines, the classifiers performance can be estimated as relative to a random sample. ROC curves convey more information than possible via a singular scalar metrix. The results confirm that using hit ratio for assessing a classifier accuracy in machine learning includes many successes that could have resulted from chance.