Editing Classification (section)

== Evaluation of accuracy ==
Unlike in [[decision theory]], it is assumed that a classifier repeats the classification task over and over. And unlike a [[lottery]], it is assumed that each classification can be either right or wrong; in the theory of measurement, classification is understood as measurement against a [[Level_of_measurement#Nominal_level|nominal]] scale. Thus it is possible to try to measure the accuracy of a classifier.

Measuring the accuracy of a classifier allows a choice to be made between two alternative classifiers. This is important both when developing a classifier and in choosing which classifier to deploy. There are however many different methods for evaluating the accuracy of a classifier and no general method for determining which method should be used in which circumstances. Different fields have taken different approaches, even in binary classification. In [[pattern recognition]], error rate is popular. The [[Gini coefficient]] and KS statistic are widely used in the credit scoring industry. [[Sensitivity and specificity]] are widely used in epidemiology and medicine. [[Precision and recall]] are widely used in information retrieval.<ref name="Hand2012">
{{Cite journal
 | author = David Hand
 | title = Assessing the Performance of Classification Methods
 | journal = [[International Statistical Review]]
 | volume = 80
 | issue = 3
 | pages = 400–414
 | year = 2012
 | doi = 10.1111/j.1751-5823.2012.00183.x
 }}</ref>

Classifier accuracy depends greatly on the characteristics of the data to be classified. There is no single classifier that works best on all given problems (a phenomenon that may be explained by the [[No free lunch in search and optimization|no-free-lunch theorem]]).