Error rate is the proportion of a system’s predictions that are incorrect. It complements success rate: together, the two rates add up to 100%. section 2.3.2
Choosing error rate as a training objective means trying to minimize incorrect predictions. For recommendations, that could mean minimizing recommendations that are not relevant to the people receiving them. section 2.3.2
A digit-recognition system
A perfect digit recognizer has a 0% error rate. A system that guesses randomly among ten digits has an average 90% error rate. In content moderation, even a small percentage of errors can produce many mislabeled examples when platforms receive large volumes of content. section 2.3.2 section 3.5









