aka pattern recognition

Supervised learning, where the output value is discrete, or categorical, or qualitative. No implicit ordering, or closeness on the variables. See video.

Many of the same methods as in regression. In fact, classification is often approached by modelling a posterior probability $p(y|x)$, and predicting the most probably $y$. Because this is a continuous function of $x$, learning it can be done by using Regression analysis

Supervised classification

Binary classification

Logistic regression (Classification with 2 classes). A simpler version: Perceptron.

Multiclass classification

Classification with $k$ classes.

Softmax regression

See Generalized linear model for other more generalized methods.

Artificial neural network (see also Deep learning)

Support vector machines. Software for SVMs: http://svmlight.joachims.org/

Nonparametric algortihm: Nearest-neighbour classification

Decision trees

• Random forest

Multiclass classification using binary classifiers

see page 16 here.

One-vs-one

One-vs-rest

May create hard learning problems when combining some classes (resulting in regions with weird shapes, that are hard to linearly separate). With decisino trees, this can be exacerbated.

Decision trees

Reducing Multiclass to Binary.E. Allwein, R. Schapire, Y. Singer

Multi-label classification

If the class labels are not mutually exclusive. This can be modelled by using multiple related binary class labels, each marking whether the input belongs to a particular class. This is a multiple output model.

Bootstrap aggregation

Generative supervised learning

Unsupervised classification: Clustering

Ordered categorical classification

Output has a notion of order, but not closeness, so it's qualitative, aka ordinal learning.

Measuring performance

using a confusion matrix

can give different costs..

(Decision theory)

TPR vs FPR

Precision vs recall

How to tune classifer to satisfy these criteria?

..

Pattern Recognition Class

https://www.wikiwand.com/en/Statistical_classification