What are Support Vector Machines?
Support Vector Machines (SVM) are algorithms based on a mathematical procedure that are used to classify or regress objects. These objects are represented by data points. For the classification, parallel support vectors are formed by certain pairs of points, which give the method its name. The support vectors with the greatest distance separate the classes from each other.
Where should Support Vector Machines be classified?
A section of the Artificial intelligence is machine learning, which is found in Supervised Learning and unsupervised learning. SVMs belong to the supervised learning algorithms, which are based on a training data set that already contains the correct classification of the data points.
This Algorithms are specialised in classifying certain objects in advance with the training data set in the learning phase. In the subsequent application phase, the algorithms can classify unknown objects. Depending on the required accuracy of the results, the learning phase can be repeated as often as desired with further training data in order to increase the proportion of correctly classified objects.
How do Support Vector Machines work?
SVMs go through the following steps in the learning phase: The training data is divided based on the classification. For both classes, the positions of the data points are identified and the division into points that are on the edge of a class and points that are surrounded by other points of the same class is made. Now the two edge points from the classes are found which have the smallest distance to each other and a straight line is drawn between them.
A perpendicular is drawn through the centre of this straight line, which is inclined around the point of intersection in such a way that it correctly divides the two classes. The maximum possible distance between the inclined perpendicular and all other boundary points of both classes is formed. This plane is called a hyperplane. The support vectors are the parallels of the hyperplane that are drawn through the edge points with the smallest distance.
How are SVMs used?
SVMs are used to classify large amounts of data. Since this procedure can develop a high computational demand, so-called slack variables can be defined to improve performance. These also serve to avoid overfitting and allow for single, incorrectly classified data points.
It is possible to make non-linear classifications in addition to linear ones.
SVMs for non-linear classification are extended with the kernel method, which can produce results by considering further dimensions. For this purpose, the training data is transferred in a higher-dimensional space until a linear hyperplane can be formed. When transferred to the original dimension, this becomes a non-linear plane that can classify objects correctly.