Statistical classification

Classification methods are methods and criteria for classification of objects or situations in classes, ie for classification. Such a process is also referred to as a classifier. Many procedures can be implemented as an algorithm; one also talks of mechanical or automatic classification. Classification methods are always application specific, so that many different methods exist.

Classification methods play a role, among others, in pattern recognition, in the artificial intelligence and documentation science or information retrieval. To evaluate a classifier can be determined in several ways.

Types of classification methods

Since a strictly hierarchical structure of classification methods is hardly possible, they can best be classified based on several properties:

  • Manual and automatic process
  • Numerical and non-numerical method
  • Statistical and distribution-free methods
  • Supervised and unsupervised methods
  • Fixed -sized and learning process
  • Parametric and non-parametric methods

Manual and automatic process

See, for example, manual indexing and automatic indexing.

In automatic mode of Klassizifierung takes place by means of an automatic process by software. The process of automatic classification can be referred to as a formal method of decision-making in new situations because of learned structures. The automated classification is a branch of machine learning.

More precisely this is the generation of an algorithm ( learning algorithm ) which - in a known and used already classified cases (data base ) - calculated structures. This newly acquired structures enable a further algorithm ( the evaluating algorithm), a new and previously unknown case, due to the observed attributes and to assign the characteristics of the known target classes.

Statistical and distribution-free methods

Statistical methods based on density calculations and probabilities, while the corresponding in distribution-free method to specify clear separation surfaces.

Supervised and unsupervised methods

Generating patterns of the existing data is also referred to as pattern recognition or discrimination supervised learning. This class divisions are specified (this can also be done by sampling). In contrast, there is not supervised learning, wherein the classes of the data are not given, but they must also be learned. However, it can in the reinforcement learning (English: reinforcement learning ) added information as to whether a classification was right or wrong. An example of unsupervised method is the cluster analysis.

Parametric and non-parametric methods

Parametric methods are based on parametric probability densities, while non-parametric methods (eg, nearest neighbor classification), the densities are usually estimated from a sample.

Examples

  • Logistic Regression
  • Square classifier
  • Abstandsklassifikator
  • Bayes classifier
  • Nearest Neighbor Classification
  • Fuzzy classifier
  • Polynomial classifier
  • Cluster method
  • Artificial neural network
  • Latent class analysis
  • Support Vector Machines

If the limits are known between the different classes in the feature space, it can be specified by a discriminant function.

Classification in machine learning Subject indexing Naive Bayes classifier Latent class model Principal component analysis Mahalanobis distance Feature vector
478434
de