Support vector machine


A support vector machine (SVN) is used to find the optimal dividing lines between classes of training data within a vector space. These can then be used to provide definitive classifications for new input data. Hence the output when a new piece of data is presented to a support vector machine is a clear classification; compare logistic regression, where the output is the probability that the data item belongs to one or the other class.

Support vector machines are based on an area of mathematics called kernel functions. As well as the standard hard-margin case where the classes are clearly clustered within a vector space, support vector machines can also map cope with the soft-margin case where there is no clear linear function dividing them. They achieve this by mapping the input vectors to higher-dimensional vectors in which the required linear demarcations do exist.

Just like logistic regression, support vector machines in their pure form can only model binary decisions between two classes. Multiclass SVN combines binary support vector machines in various ways to model multiple choice sets in a similar fashion to multinomial logistic regression.

Hard-margin support vector machine Soft-margin support vector machine Multiclass SVN
has functional building block
has input data type
IDT_Vector of quantitative variables
has internal model
has output data type
has learning style
has parametricity
has relevance
sometimes supports
mathematically similar to
SPT_Elastic net