Pattern recognition is the assignment of observations to class labels or structured descriptions using measured features. Here "category" is ordinary English
for a kind of observation; it does not denote a category
in category theory. In statistical pattern recognition,
the feature vector is modeled as a random variable 
, and a classifier is a function
that assigns an observed value 
to one of the classes 
, ..., 
.
A Bayes classifier is the classifier obtained from the Bayes decision rule. Given the posterior
probabilities 
and a loss function 
, it chooses an action 
that minimizes the conditional risk
 |
For a zero-one loss function, an incorrect label costs 1 and a correct label costs 0, so this reduces to choosing the most probable class,
 |
Pattern recognition includes classification, cluster analysis, feature extraction, and sequence labeling.
