Statistical classification categorizes data into groups using statistical methods and algorithms. A computer program sorts individual observations by analyzing quantifiable properties called features. The term classifier refers to the algorithm itself or the mathematical function it implements.
R. Gnanadesikan published Methods for Statistical Data Analysis of Multivariate Observations in 1977. Ronald Fisher developed linear discriminant functions for two-group problems earlier than this publication date. C.R. Rao wrote Advanced Statistical Methods in Multivariate Analysis in 1952.
Bayesian procedures incorporate information about relative group sizes naturally while Frequentist methods rely on observed frequencies without prior population estimates. Markov chain Monte Carlo computations now make Bayesian calculations feasible where approximations were necessary before this development. Group-membership probabilities provide more informative outcomes than single labels.
S. Har-Peled published Constraint Classification for Multiclass Classification and Ranking in 2003. D. Roth and D. Zimak contributed to the same proceedings at MIT Press. Binary classification involves only two classes while multiclass handles several categories.
Individual instances use feature vectors of measurable properties for prediction such as pixels or word occurrence frequencies. Pixels represent features in image analysis contexts while word counts serve as integer values in email systems. Discrete algorithms require discretizing real or integer data into groups like ranges less than five, between five and ten, or greater than ten.