View source for Cluster analysis

{{Short description|Grouping a set of objects by similarity}}
[[File:Cluster-2.svg|thumb|The result of a cluster analysis shown as the coloring of the squares into three clusters]]
{{Machine learning}}

'''Cluster analysis''' or '''clustering''' is the task of grouping a set of objects in such a way that objects in the same group (called a '''cluster''') are more [[Similarity measure|similar]] (in some specific sense defined by the analyst) to each other than to those in other groups (clusters). It is a main task of [[Exploratory data analysis|exploratory data analysis]], and a common technique for [[Statistics|statistical]] [[Data analysis|data analysis]], used in many fields, including [[Pattern recognition|pattern recognition]], [[Image analysis|image analysis]], [[Information retrieval|information retrieval]], [[Biology:Bioinformatics|bioinformatics]], [[Data compression|data compression]], [[Computer graphics|computer graphics]] and [[Machine learning|machine learning]].

Cluster analysis refers to a family of algorithms and tasks rather than one specific [[Algorithm|algorithm]]. It can be achieved by various algorithms that differ significantly in their understanding of what constitutes a cluster and how to efficiently find them. Popular notions of clusters include groups with small distances between cluster members, dense areas of the data space, intervals or particular statistical distributions. Clustering can therefore be formulated as a [[Multi-objective optimization|multi-objective optimization]] problem. The appropriate clustering algorithm and parameter settings (including parameters such as the [[Metric (mathematics)|distance function]] to use, a density threshold or the number of expected clusters) depend on the individual [[Data set|data set]] and intended use of the results. Cluster analysis as such is not an automatic task, but an iterative process of knowledge discovery or interactive multi-objective optimization that involves trial and failure. It is often necessary to modify data preprocessing and model parameters until the result achieves the desired properties.

Besides the term ''clustering'', there is a number of terms with similar meanings, including ''automatic [[Statistical classification|classification]]'', ''[[Numerical taxonomy|numerical taxonomy]]'', ''botryology'' (from Greek βότρυς "grape"), ''typological analysis'', and ''[[Community structure|community detection]]''. The subtle differences are often in the use of the results: while in data mining, the resulting groups are the matter of interest, in automatic classification the resulting discriminative power is of interest.

Cluster analysis was originated in anthropology by Driver and Kroeber in 1932<ref>{{Cite journal|last=Driver and Kroeber|date=1932|title=Quantitative Expression of Cultural Relationships|url=http://dpg.lib.berkeley.edu/webdb/anthpubs/search?all=&volume=31&journal=1&item=5|journal=University of California Publications in American Archaeology and Ethnology|volume=Quantitative Expression of Cultural Relationships|pages=211–256|publisher=University of California Press|publication-place=Berkeley, CA|publication-date=1932|access-date=2019-02-18|archive-date=2020-12-06|archive-url=https://web.archive.org/web/20201206053117/https://dpg.lib.berkeley.edu/webdb/anthpubs/search?all=&volume=31&journal=1&item=5|url-status=dead}}</ref> and introduced to psychology by Joseph Zubin in 1938<ref>{{Cite journal|last=Zubin|first=Joseph|date=1938|title=A technique for measuring like-mindedness.|journal=The Journal of Abnormal and Social Psychology|language=en|volume=33|issue=4|pages=508–516|doi=10.1037/h0055441|issn=0096-851X}}</ref> and Robert Tryon in 1939<ref>{{cite book | title = Cluster Analysis: Correlation Profile and Orthometric (factor) Analysis for the Isolation of Unities in Mind and Personality | first = Robert C. | last = Tryon |  publisher = Edwards Brothers | year = 1939}}</ref> and famously used by Cattell beginning in 1943<ref>{{cite journal | last = Cattell | first = R. B. | date = 1943 | title = The description of personality: Basic traits resolved into clusters | journal = Journal of Abnormal and Social Psychology | volume = 38 | issue = 4 | pages = 476–506 | doi=10.1037/h0054116}}</ref> for trait theory classification in [[Social:Personality psychology|personality psychology]].

== Definition ==

The notion of a "cluster" cannot be precisely defined, which is one of the reasons why there are so many clustering algorithms.<ref name="estivill">{{cite journal | title=Why so many clustering algorithms – A Position Paper | last = Estivill-Castro | first = Vladimir | s2cid = 7329935 | journal=ACM SIGKDD Explorations Newsletter |date=20 June 2002 | volume= 4 | issue=1 | pages=65–75 | doi=10.1145/568574.568575}}</ref> There is a common denominator: a group of data objects. However, different researchers employ different cluster models, and for each of these cluster models again different algorithms can be given. The notion of a cluster, as found by different algorithms, varies significantly in its properties. Understanding these "cluster models" is key to understanding the differences between the various algorithms. Typical cluster models include:
* ''{{vanchor|Connectivity model}}s'': for example, [[Hierarchical clustering|hierarchical clustering]] builds models based on distance connectivity.
* ''{{vanchor|Centroid model}}s'': for example, the k-means algorithm represents each cluster by a single mean vector.
* ''{{vanchor|Distribution model}}s'': clusters are modeled using statistical distributions, such as [[Multivariate normal distribution|multivariate normal distribution]]s used by the expectation-maximization algorithm.
* ''{{vanchor|Density model}}s'': for example, [[DBSCAN]] and OPTICS defines clusters as connected dense regions in the data space.
* ''{{vanchor|Subspace model}}s'': in [[Biclustering|biclustering]] (also known as co-clustering or two-mode-clustering), clusters are modeled with both cluster members and relevant attributes.
* ''{{vanchor|Group model}}s'': some algorithms do not provide a refined model for their results and just provide the grouping information.
* ''{{vanchor|Graph-based model}}s'': a [[Clique (graph theory)|clique]], that is, a subset of nodes in a [[Graph (discrete mathematics)|graph]] such that every two nodes in the subset are connected by an edge can be considered as a prototypical form of cluster. Relaxations of the complete connectivity requirement (a fraction of the edges can be missing) are known as quasi-cliques, as in the [[HCS clustering algorithm]].
* ''Signed graph models'': Every [[Path (graph theory)|path]] in a [[Signed graph|signed graph]] has a [[Sign (mathematics)|sign]] from the product of the signs on the edges. Under the assumptions of [[Philosophy:Balance theory|balance theory]], edges may change sign and result in a bifurcated graph. The weaker "clusterability axiom" (no [[Cycle (graph theory)|cycle]] has exactly one negative edge) yields results with more than two clusters, or subgraphs with only positive edges.<ref>James A. Davis (May 1967) "Clustering and structural balance in graphs", Human Relations 20:181–7</ref>
* ''{{vanchor|Neural model}}s'': the most well known [[Unsupervised learning|unsupervised]] [[Neural network|neural network]] is the [[Self-organizing map|self-organizing map]] and these models can usually be characterized as similar to one or more of the above models, and including subspace models when neural networks implement a form of Principal Component Analysis or Independent Component Analysis.

A "clustering" is essentially a set of such clusters, usually containing all objects in the data set. Additionally, it may specify the relationship of the clusters to each other, for example, a hierarchy of clusters embedded in each other. Clusterings can be roughly distinguished as:
* ''{{vanchor|Hard clustering}}'': each object belongs to a cluster or not
* ''{{vanchor|Soft clustering}}'' (also: ''{{vanchor|[[Fuzzy clustering|fuzzy clustering]]}}''): each object belongs to each cluster to a certain degree (for example, a likelihood of belonging to the cluster)

There are also finer distinctions possible, for example:

* ''{{vanchor|Strict partitioning clustering}}'': each object belongs to exactly one cluster
* ''{{vanchor|Strict partitioning clustering with outliers}}'': objects can also belong to no cluster, in which case they are considered [[Anomaly detection|outliers]]
* ''{{vanchor|Overlapping clustering}}'' (also: ''alternative clustering'', ''multi-view clustering''): objects may belong to more than one cluster; usually involving hard clusters
* ''{{vanchor|Hierarchical clustering}}'': objects that belong to a child cluster also belong to the parent cluster
* ''{{vanchor|Subspace clustering}}'': while an overlapping clustering, within a uniquely defined subspace, clusters are not expected to overlap

== Algorithms ==

As listed above, clustering algorithms can be categorized based on their cluster model. The following overview will only list the most prominent examples of clustering algorithms, as there are possibly over 100 published clustering algorithms. Not all provide models for their clusters and can thus not easily be categorized. An overview of algorithms explained in Wikipedia can be found in the [[List of algorithms#Statistics|list of statistics algorithms]].

There is no objectively "correct" clustering algorithm, but as it was noted, "clustering is in the eye of the beholder."<ref name="estivill" /> The most appropriate clustering algorithm for a particular problem often needs to be chosen experimentally, unless there is a mathematical reason to prefer one cluster model over another. An algorithm that is designed for one kind of model will generally fail on a data set that contains a radically different kind of model.<ref name="estivill" /> For example, k-means cannot find non-convex clusters.<ref name="estivill" /> Most traditional clustering methods assume the clusters exhibit a spherical, elliptical or convex shape.<ref>{{Cite journal |last=Gao |first=Caroline X. |last2=Dwyer |first2=Dominic |last3=Zhu |first3=Ye |last4=Smith |first4=Catherine L. |last5=Du |first5=Lan |last6=Filia |first6=Kate M. |last7=Bayer |first7=Johanna |last8=Menssink |first8=Jana M. |last9=Wang |first9=Teresa |last10=Bergmeir |first10=Christoph |last11=Wood |first11=Stephen |last12=Cotton |first12=Sue M. |date=2023-09-01 |title=An overview of clustering methods with guidelines for application in mental health research |url=https://www.sciencedirect.com/science/article/pii/S0165178123002159 |journal=Psychiatry Research |volume=327 |pages=115265 |doi=10.1016/j.psychres.2023.115265 |issn=0165-1781|doi-access=free |hdl=10481/84538 |hdl-access=free }}</ref>

=== Connectivity-based clustering (hierarchical clustering) ===
{{Main|Hierarchical clustering}}
Connectivity-based clustering, also known as ''[[Hierarchical clustering|hierarchical clustering]]'', is based on the core idea of objects being more related to nearby objects than to objects farther away. These algorithms connect "objects" to form "clusters" based on their distance. A cluster can be described largely by the maximum distance needed to connect parts of the cluster. At different distances, different clusters will form, which can be represented using a [[Dendrogram|dendrogram]], which explains where the common name "[[Hierarchical clustering|hierarchical clustering]]" comes from: these algorithms do not provide a single partitioning of the data set, but instead provide an extensive hierarchy of clusters that merge with each other at certain distances. In a dendrogram, the y-axis marks the distance at which the clusters merge, while the objects are placed along the x-axis such that the clusters don't mix.

Connectivity-based clustering is a whole family of methods that differ by the way distances are computed. Apart from the usual choice of distance functions, the user also needs to decide on the linkage criterion (since a cluster consists of multiple objects, there are multiple candidates to compute the distance) to use. Popular choices are known as [[Single-linkage clustering|single-linkage clustering]] (the minimum of object distances), complete linkage clustering (the maximum of object distances), and [[Biology:UPGMA|UPGMA]] or [[WPGMA]] ("Unweighted or Weighted Pair Group Method with Arithmetic Mean", also known as average linkage clustering). Furthermore, hierarchical clustering can be agglomerative (starting with single elements and aggregating them into clusters) or divisive (starting with the complete data set and dividing it into partitions).

These methods will not produce a unique partitioning of the data set, but a hierarchy from which the user still needs to choose appropriate clusters. They are not very robust towards outliers, which will either show up as additional clusters or even cause other clusters to merge (known as "chaining phenomenon", in particular with [[Single-linkage clustering|single-linkage clustering]]). In the general case, the complexity is <math>\mathcal{O}(n^3)</math> for agglomerative clustering and <math>\mathcal{O}(2^{n-1})</math> for divisive clustering,<ref>{{cite book | last = Everitt | first = Brian | title = Cluster analysis | publisher = Wiley | location = Chichester, West Sussex, U.K | year = 2011 | isbn = 9780470749913 }}</ref> which makes them too slow for large data sets. For some special cases, optimal efficient methods (of complexity <math>\mathcal{O}(n^2)</math>) are known: SLINK<ref>{{cite journal | first = R. | last = Sibson | title=SLINK: an optimally efficient algorithm for the single-link cluster method | journal=The Computer Journal | volume=16 | issue=1 | pages=30–34 | year=1973 | publisher=British Computer Society | url=http://www.cs.gsu.edu/~wkim/index_files/papers/sibson.pdf | doi=10.1093/comjnl/16.1.30 }}</ref> for single-linkage and CLINK<ref>{{cite journal | first = D. | last = Defays | title=An efficient algorithm for a complete link method | journal=The Computer Journal | volume=20 | issue=4 | pages=364–366 | year=1977 | publisher=British Computer Society | doi=10.1093/comjnl/20.4.364}}</ref> for complete-linkage clustering.

<gallery caption="Linkage clustering examples" widths="200px" heights="200px">
File:SLINK-Gaussian-data.svg|Single-linkage on Gaussian data. At 35 clusters, the biggest cluster starts fragmenting into smaller parts, while before it was still connected to the second largest due to the single-link effect.
File:SLINK-density-data.svg|Single-linkage on density-based clusters. 20 clusters extracted, most of which contain single elements, since linkage clustering does not have a notion of "noise".
</gallery>

=== Centroid-based clustering ===
{{Main|K-means clustering}}
In centroid-based clustering, each cluster is represented by a central vector, which is not necessarily a member of the data set. When the number of clusters is fixed to ''k'', [[K-means clustering|''k''-means clustering]] gives a formal definition as an optimization problem: find the ''k'' cluster centers and assign the objects to the nearest cluster center, such that the squared distances from the cluster are minimized.

The optimization problem itself is known to be NP-hard, and thus the common approach is to search only for approximate solutions. A particularly well known approximate method is [[Lloyd's algorithm]],<ref name="lloyd">{{Cite journal | last1 = Lloyd | first1 = S. | title = Least squares quantization in PCM | doi = 10.1109/TIT.1982.1056489 | journal = IEEE Transactions on Information Theory | volume = 28 | issue = 2 | pages = 129–137 | year = 1982 | s2cid = 10833328 }}</ref> often just referred to as "''k-means algorithm''" (although [[K-means clustering#History|another algorithm introduced this name]]). It does however only find a [[Local optimum|local optimum]], and is commonly run multiple times with different random initializations. Variations of ''k''-means often include such optimizations as choosing the best of multiple runs, but also restricting the centroids to members of the data set ([[K-medoids|''k''-medoids]]), choosing [[Median|median]]s ([[K-medians clustering|''k''-medians clustering]]), choosing the initial centers less randomly ([[K-means++|''k''-means++]]) or allowing a fuzzy cluster assignment ([[Fuzzy clustering|fuzzy c-means]]).

Most ''k''-means-type algorithms require the number of clusters – ''k'' – to be specified in advance, which is considered to be one of the biggest drawbacks of these algorithms. Furthermore, the algorithms prefer clusters of approximately similar size, as they will always assign an object to the nearest centroid. This often leads to incorrectly cut borders of clusters (which is not surprising since the algorithm optimizes cluster centers, not cluster borders).

K-means has a number of interesting theoretical properties. First, it partitions the data space into a structure known as a [[Voronoi diagram]]. Second, it is conceptually close to nearest neighbor classification, and as such is popular in [[Machine learning|machine learning]]. Third, it can be seen as a variation of model based clustering, and Lloyd's algorithm as a variation of the Expectation-maximization algorithm for this model discussed below.

<gallery widths="200" heights="200" caption="''k''-means clustering examples">
File:KMeans-Gaussian-data.svg|''k''-means separates data into Voronoi cells, which assumes equal-sized clusters (not adequate here).
File:KMeans-density-data.svg|''k''-means cannot represent density-based clusters.
</gallery>

Centroid-based clustering problems such as ''k''-means and ''k''-medoids are special cases of the uncapacitated, metric [[Optimal facility location|facility location problem]], a canonical problem in the operations research and computational geometry communities. In a basic facility location problem (of which there are numerous variants that model more elaborate settings), the task is to find the best warehouse locations to optimally service a given set of consumers. One may view "warehouses" as cluster centroids and "consumer locations" as the data to be clustered. This makes it possible to apply the well-developed algorithmic solutions from the facility location literature to the presently considered centroid-based clustering problem.

=== Distribution-based clustering ===
The clustering model most closely related to statistics is based on [[Probability distribution|distribution models]]. Clusters can then easily be defined as objects belonging most likely to the same distribution. A convenient property of this approach is that this closely resembles the way artificial data sets are generated: by sampling random objects from a distribution.

While the theoretical foundation of these methods is excellent, they suffer from [[Overfitting|overfitting]] unless constraints are put on the model complexity. A more complex model will usually be able to explain the data better, which makes choosing the appropriate model complexity inherently difficult.

One prominent method is known as Gaussian mixture models (using the expectation-maximization algorithm). Here, the data set is usually modeled with a fixed (to avoid overfitting) number of Gaussian distributions that are initialized randomly and whose parameters are iteratively optimized to better fit the data set. This will converge to a [[Local optimum|local optimum]], so multiple runs may produce different results. In order to obtain a hard clustering, objects are often then assigned to the Gaussian distribution they most likely belong to; for soft clusterings, this is not necessary.

Distribution-based clustering produces complex models for clusters that can capture [[Correlation and dependence|correlation and dependence]] between attributes. However, these algorithms put an extra burden on the user: for many real data sets, there may be no concisely defined mathematical model (e.g. assuming Gaussian distributions is a rather strong assumption on the data).

<gallery widths="200" heights="200" caption="Gaussian mixture model clustering examples">
File:EM-Gaussian-data.svg|On Gaussian-distributed data, <abbr title="expectation–maximization">EM</abbr> works well, since it uses Gaussians for modelling clusters.
File:EM-density-data.svg|Density-based clusters cannot be modeled using Gaussian distributions.
</gallery>

=== Density-based clustering ===
In density-based clustering,<ref>{{Cite journal
 | title = Density-based Clustering
 | journal = WIREs Data Mining and Knowledge Discovery
 | volume = 1
 | issue = 3
 | year = 2011
 | pages = 231–240
 | url = http://wires.wiley.com/WileyCDA/WiresArticle/wisId-WIDM30.html
 | doi = 10.1002/widm.30
| s2cid = 36920706 }}</ref> clusters are defined as areas of higher density than the remainder of the data set. Objects in sparse areas – that are required to separate clusters – are usually considered to be noise and border points.

The most popular<ref>[http://academic.research.microsoft.com/CSDirectory/paper_category_7.htm Microsoft academic search: most cited data mining articles] {{webarchive|url=https://web.archive.org/web/20100421170848/http://academic.research.microsoft.com/CSDirectory/Paper_category_7.htm |date=2010-04-21 }}: DBSCAN is on rank 24, when accessed on: 4/18/2010</ref> density based clustering method is [[DBSCAN]].<ref>{{Cite conference | author1-first = Martin | author1-last = Ester |  author2-first = Hans-Peter | author2-last = Kriegel | author3-first = Jörg | author3-last = Sander | author4-first = Xiaowei | author4-last = Xu | title = A density-based algorithm for discovering clusters in large spatial databases with noise | pages = 226–231 | editor1-first = Evangelos | editor1-last = Simoudis | editor2-first = Jiawei | editor2-last = Han | editor3-first = Usama M. | editor3-last = Fayyad | book-title = Proceedings of the Second International Conference on Knowledge Discovery and Data Mining (KDD-96) | publisher = AAAI Press | year = 1996 | isbn = 1-57735-004-9 }}</ref> In contrast to many newer methods, it features a well-defined cluster model called "density-reachability". Similar to linkage based clustering, it is based on connecting points within certain distance thresholds. However, it only connects points that satisfy a density criterion, in the original variant defined as a minimum number of other objects within this radius. A cluster consists of all density-connected objects (which can form a cluster of an arbitrary shape, in contrast to many other methods) plus all objects that are within these objects' range. Another interesting property of DBSCAN is that its complexity is fairly low – it requires a linear number of range queries on the database – and that it will discover essentially the same results (it is [[Deterministic algorithm|deterministic]] for core and noise points, but not for border points) in each run, therefore there is no need to run it multiple times. [[OPTICS algorithm|OPTICS]]<ref>{{Cite conference
 | title = OPTICS: Ordering Points To Identify the Clustering Structure
 | year = 1999
 | pages = 49–60
 | book-title = ACM SIGMOD international conference on Management of data
 | publisher = ACM Press
 | citeseerx = 10.1.1.129.6542
}}</ref> is a generalization of DBSCAN that removes the need to choose an appropriate value for the range parameter <math>\varepsilon</math>, and produces a hierarchical result related to that of [[Hierarchical clustering|linkage clustering]]. DeLi-Clu,<ref name="ReferenceA">{{Cite conference| doi = 10.1007/11731139_16| isbn = 978-3-540-33206-0| contribution = DeLi-Clu: Boosting Robustness, Completeness, Usability, and Efficiency of Hierarchical Clustering by a Closest Pair Ranking| year = 2006| last1 = Achtert | first1 = E.| last2 = Böhm| series = Lecture Notes in Computer Science | first2 = C.| last3 = Kröger | first3 = P.| pages = 119–128|title= Advances in Knowledge Discovery and Data Mining| volume = 3918| citeseerx = 10.1.1.64.1161}}</ref> Density-Link-Clustering combines ideas from [[Single-linkage clustering|single-linkage clustering]] and OPTICS, eliminating the <math>\varepsilon</math> parameter entirely and offering performance improvements over OPTICS by using an R-tree index.

The key drawback of [[DBSCAN]] and OPTICS is that they expect some kind of density drop to detect cluster borders. On data sets with, for example, overlapping Gaussian distributions – a common use case in artificial data – the cluster borders produced by these algorithms will often look arbitrary, because the cluster density decreases continuously. On a data set consisting of mixtures of Gaussians, these algorithms are nearly always outperformed by methods such as [[Expectation–maximization algorithm|EM clustering]] that are able to precisely model this kind of data.

Mean-shift is a clustering approach where each object is moved to the densest area in its vicinity, based on [[Kernel density estimation|kernel density estimation]]. Eventually, objects converge to local maxima of density. Similar to k-means clustering, these "density attractors" can serve as representatives for the data set, but mean-shift can detect arbitrary-shaped clusters similar to DBSCAN. Due to the expensive iterative procedure and density estimation, mean-shift is usually slower than DBSCAN or k-Means. Besides that, the applicability of the mean-shift algorithm to multidimensional data is hindered by the unsmooth behaviour of the kernel density estimate, which results in over-fragmentation of cluster tails.<ref name="ReferenceA"/>

<gallery widths="200" heights="200" caption="Density-based clustering examples">
File:DBSCAN-density-data.svg|Density-based clustering with [[DBSCAN]]
File:DBSCAN-Gaussian-data.svg|[[DBSCAN]] assumes clusters of similar density, and may have problems separating nearby clusters.
File:OPTICS-Gaussian-data.svg|[[OPTICS algorithm|OPTICS]] is a DBSCAN variant, improving handling of different densities clusters.
</gallery>

=== Grid-based clustering ===
The grid-based technique is used for a [[Multidimensional scaling|multi-dimensional]] data set.<ref>{{Cite book|editor-last1=Aggarwal|editor-first1=Charu C.|editor-last2=Reddy|editor-first2=Chandan K.|title=Data Clustering : Algorithms and Applications|isbn=978-1-315-37351-5|oclc=1110589522}}</ref> In this technique, we create a grid structure, and the comparison is performed on grids (also known as cells). The grid-based technique is fast and has low computational complexity. There are two types of grid-based clustering methods: STING and CLIQUE. Steps involved in grid-based clustering [[Algorithm|algorithm]] are:
# Divide data space into a finite number of cells.
# Randomly select a cell ‘c’, where c should not be traversed beforehand.
# Calculate the density of ‘c’
# If the density of ‘c’ greater than threshold density
## Mark cell ‘c’ as a new cluster
## Calculate the density of all the neighbors of ‘c’
## If the density of a neighboring cell is greater than threshold density then, add the cell in the cluster and repeat steps 4.2 and 4.3 till there is no neighbor with a density greater than threshold density.
# Repeat steps 2,3 and 4 till all the cells are traversed.
# Stop.

=== Recent developments ===
In recent years, considerable effort has been put into improving the performance of existing algorithms.<ref>{{cite conference | first = D. | last = Sculley |title=Web-scale k-means clustering |conference=Proc. 19th WWW |year=2010}}</ref><ref>{{cite journal | last1 = Huang | first1 = Z. | s2cid = 11323096 | year = 1998 | title = Extensions to the ''k''-means algorithm for clustering large data sets with categorical values | journal = Data Mining and Knowledge Discovery | volume = 2 | issue = 3| pages = 283–304 | doi = 10.1023/A:1009769707641 }}</ref> Among them are ''CLARANS'',<ref>R. Ng and J. Han. "Efficient and effective clustering method for spatial data mining". In: Proceedings of the 20th VLDB Conference, pages 144–155, Santiago, Chile, 1994.</ref> and ''BIRCH''.<ref>Tian Zhang, Raghu Ramakrishnan, Miron Livny. "[http://www.cs.du.edu/~leut/4423/papers/zhang96birch.pdf An Efficient Data Clustering Method for Very Large Databases]." In: Proc. Int'l Conf. on Management of Data, ACM SIGMOD, pp. 103–114.</ref> With the recent need to process larger and larger data sets (also known as [[Big data|big data]]), the willingness to trade semantic meaning of the generated clusters for performance has been increasing. This led to the development of pre-clustering methods such as [[Canopy clustering algorithm|canopy clustering]], which can process huge data sets efficiently, but the resulting "clusters" are merely a rough pre-partitioning of the data set to then analyze the partitions with existing slower methods such as [[K-means clustering|k-means clustering]].

For high-dimensional data, many of the existing methods fail due to the [[Curse of dimensionality|curse of dimensionality]], which renders particular distance functions problematic in high-dimensional spaces. This led to new [[Clustering high-dimensional data|clustering algorithms for high-dimensional data]] that focus on subspace clustering (where only some attributes are used, and cluster models include the relevant attributes for the cluster) and [[Correlation clustering|correlation clustering]] that also looks for arbitrary rotated ("correlated") subspace clusters that can be modeled by giving a [[Correlation|correlation]] of their attributes.<ref>{{cite journal|last1=Kriegel|first1=Hans-Peter|last2=Kröger|first2=Peer|last3=Zimek|first3=Arthur|author-link3=Arthur Zimek|title=Subspace clustering|journal=Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery|date=July 2012|volume=2|issue=4|pages=351–364|doi=10.1002/widm.1057|s2cid=7241355}}</ref> Examples for such clustering algorithms are CLIQUE<ref>{{Cite journal | last1 = Agrawal | first1 = R. | last2 = Gehrke | first2 = J. | last3 = Gunopulos | first3 = D. | last4 = Raghavan | first4 = P. | title = Automatic Subspace Clustering of High Dimensional Data | doi = 10.1007/s10618-005-1396-1 | journal = Data Mining and Knowledge Discovery | volume = 11 | pages = 5–33 | year = 2005 | citeseerx = 10.1.1.131.5152 | s2cid = 9289572 }}</ref> and [[SUBCLU]].<ref>Karin Kailing, [[Biography:Hans-Peter Kriegel|Hans-Peter Kriegel]] and Peer Kröger. ''Density-Connected Subspace Clustering for High-Dimensional Data''. In: ''Proc. SIAM Int. Conf. on Data Mining (SDM'04)'', pp. 246–257, 2004.</ref>

Ideas from density-based clustering methods (in particular the [[DBSCAN]]/OPTICS family of algorithms) have been adapted to subspace clustering (HiSC,<ref>{{Cite conference| doi = 10.1007/11871637_42| contribution = Finding Hierarchies of Subspace Clusters| isbn = 978-3-540-45374-1| year = 2006| last1 = Achtert | first1 = E.| last2 = Böhm | first2 = C.| last3 = Kriegel | first3 = H.-P.| series = Lecture Notes in Computer Science |  last4 = Kröger | first4 = P.| last5 = Müller-Gorman | first5 = I.| last6 = Zimek | first6 = A.|author-link6=Arthur Zimek| pages = 446–453| title= Knowledge Discovery in Databases: PKDD 2006| volume = 4213| citeseerx = 10.1.1.705.2956}}</ref> hierarchical subspace clustering and DiSH<ref>{{Cite conference| doi = 10.1007/978-3-540-71703-4_15| contribution = Detection and Visualization of Subspace Cluster Hierarchies| isbn = 978-3-540-71702-7| year = 2007| last1 = Achtert | first1 = E.| last2 = Böhm | first2 = C.| last3 = Kriegel | first3 = H. P. | author-link3 =Hans-Peter Kriegel| series = Lecture Notes in Computer Science| last4 = Kröger | first4 = P.| last5 = Müller-Gorman | first5 = I.| last6 = Zimek | first6 = A.|author-link6=Arthur Zimek| volume = 4443| pages = 152–163| title= Advances in Databases: Concepts, Systems and Applications| citeseerx = 10.1.1.70.7843}}</ref>) and correlation clustering (HiCO,<ref>{{Cite book| doi = 10.1109/SSDBM.2006.35| isbn = 978-0-7695-2590-7| year = 2006| last1 = Achtert | first1 = E.| last2 = Böhm | first2 = C.| last3 = Kröger | first3 = P.| last4 = Zimek | first4 = A.| title = 18th International Conference on Scientific and Statistical Database Management (SSDBM'06)| chapter = Mining Hierarchies of Correlation Clusters|author-link4=Arthur Zimek| pages = 119–128| citeseerx = 10.1.1.707.7872| s2cid = 2679909}}</ref> hierarchical correlation clustering, 4C<ref>{{Cite book | last1 = Böhm | first1 = C. | last2 = Kailing | first2 = K. | last3 = Kröger | first3 = P. | last4 = Zimek | first4 = A. |author-link4=Arthur Zimek| chapter = Computing Clusters of Correlation Connected objects | doi = 10.1145/1007568.1007620 | title = Proceedings of the 2004 ACM SIGMOD international conference on Management of data - SIGMOD '04 | pages = 455 | year = 2004 | isbn = 978-1581138597 | citeseerx = 10.1.1.5.1279 | s2cid = 6411037 }}</ref> using "correlation connectivity" and ERiC<ref>{{Cite book | last1 = Achtert | first1 = E. | last2 = Bohm | first2 = C. | last3 = Kriegel | first3 = H. P.  | author-link3=Hans-Peter Kriegel| last4 = Kröger | first4 = P. | last5 = Zimek | first5 = A. |author-link5=Arthur Zimek| doi = 10.1109/SSDBM.2007.21 | chapter = On Exploring Complex Relationships of Correlation Clusters | title = 19th International Conference on Scientific and Statistical Database Management (SSDBM 2007) | pages = 7 | year = 2007 | isbn = 978-0-7695-2868-7 | citeseerx = 10.1.1.71.5021 | s2cid = 1554722 }}</ref> exploring hierarchical density-based correlation clusters).

Several different clustering systems based on [[Mutual information|mutual information]] have been proposed. One is Marina Meilă's ''[[Variation of information|variation of information]]'' metric;<ref>{{Cite conference|last=Meilă|first=Marina | contribution=Comparing Clusterings by the Variation of Information |title=Learning Theory and Kernel Machines|year=2003|pages=173–187|doi=10.1007/978-3-540-45167-9_14|series=Lecture Notes in Computer Science|isbn=978-3-540-40720-1|volume=2777}}</ref> another provides hierarchical clustering.<ref>{{cite journal| first1 = Alexander | last1 = Kraskov | first2 = Harald | last2 = Stögbauer | first3 = Ralph G. | last3 = Andrzejak | first4 = Peter | last4 = Grassberger | title = Hierarchical Clustering Based on Mutual Information | date = 1 December 2003 | arxiv= q-bio/0311039| bibcode = 2003q.bio....11039K }}</ref> Using genetic algorithms, a wide range of different fit-functions can be optimized, including mutual information.<ref>{{cite journal | last = Auffarth | first = B. | title = Clustering by a Genetic Algorithm with Biased Mutation Operator | journal = Wcci Cec | publisher = IEEE | date = July 18–23, 2010 |url=http://www.diva-portal.org/smash/get/diva2:456742/FULLTEXT02}}</ref> Also [[Belief propagation|belief propagation]], a recent development in [[Computer science|computer science]] and [[Physics:Statistical physics|statistical physics]], has led to the creation of new types of clustering algorithms.<ref>{{Cite journal| first1 =  B. J. | last1 = Frey | first2 = D. | last2 = Dueck| title=Clustering by Passing Messages Between Data Points|journal=Science| volume=315| pages=972–976|doi= 10.1126/science.1136800|year= 2007|issue= 5814|pmid= 17218491 | bibcode = 2007Sci...315..972F | citeseerx = 10.1.1.121.3145 | s2cid = 6502291 }}</ref>

== Evaluation and assessment ==
Evaluation (or "validation") of clustering results is as difficult as the clustering itself.<ref name="pfitzner">{{cite journal|last1=Pfitzner|first1=Darius|last2=Leibbrandt|first2=Richard|last3=Powers|first3=David|s2cid=6935380|date=2009|title=Characterization and evaluation of similarity measures for pairs of clusterings|journal=Knowledge and Information Systems|publisher=Springer|volume=19|issue=3|pages=361–394|doi=10.1007/s10115-008-0150-6}}</ref> Popular approaches involve "''internal''" evaluation, where the clustering is summarized to a single quality score, "''external''" evaluation, where the clustering is compared to an existing "ground truth" classification, "''manual''" evaluation by a human expert, and "''indirect''" evaluation by evaluating the utility of the clustering in its intended application.<ref name=":0" />

Internal evaluation measures suffer from the problem that they represent functions that themselves can be seen as a clustering objective. For example, one could cluster the data set by the Silhouette coefficient; except that there is no known efficient algorithm for this. By using such an internal measure for evaluation, one rather compares the similarity of the optimization problems,<ref name=":0" /> and not necessarily how useful the clustering is.

External evaluation has similar problems: if we have such "ground truth" labels, then we would not need to cluster; and in practical applications we usually do not have such labels. On the other hand, the labels only reflect one possible partitioning of the data set, which does not imply that there does not exist a different, and maybe even better, clustering.

Neither of these approaches can therefore ultimately judge the actual quality of a clustering, but this needs human evaluation,<ref name=":0">{{Cite book|title=The Text Mining Handbook: Advanced Approaches in Analyzing Unstructured Data|last1=Feldman|first1=Ronen|last2=Sanger|first2=James|date=2007-01-01|publisher=Cambridge Univ. Press|isbn=978-0521836579|oclc=915286380}}</ref> which is highly subjective. Nevertheless, such statistics can be quite informative in identifying bad clusterings,<ref name=":1">{{Cite book|title=Text Mining: Predictive Methods for Analyzing Unstructured Information|last1=Weiss|first1=Sholom M.|last2=Indurkhya|first2=Nitin|last3=Zhang|first3=Tong|last4=Damerau|first4=Fred J.|publisher=Springer|year=2005|isbn=978-0387954332|oclc=803401334}}</ref> but one should not dismiss subjective human evaluation.<ref name=":1" />

=== Internal evaluation ===
When a clustering result is evaluated based on the data that was clustered itself, this is called internal evaluation. These methods usually assign the best score to the algorithm that produces clusters with high similarity within a cluster and low similarity between clusters. One drawback of using internal criteria in cluster evaluation is that high scores on an internal measure do not necessarily result in effective information retrieval applications.<ref name="Christopher D. Manning, Prabhakar Raghavan & Hinrich Schutze">{{Cite book
 | first1 = Christopher D. | last1 = Manning | first2 = Prabhakar | last2 = Raghavan | first3 = Hinrich | last3 = Schütze
 | title = Introduction to Information Retrieval
 | publisher = Cambridge University Press
 | isbn = 978-0-521-86571-5
| date = 2008-07-07 }}</ref> Additionally, this evaluation is biased towards algorithms that use the same cluster model. For example, k-means clustering naturally optimizes object distances, and a distance-based internal criterion will likely overrate the resulting clustering.

Therefore, the internal evaluation measures are best suited to get some insight into situations where one algorithm performs better than another, but this shall not imply that one algorithm produces more valid results than another.<ref name="estivill" /> Validity as measured by such an index depends on the claim that this kind of structure exists in the data set. An algorithm designed for some kind of models has no chance if the data set contains a radically different set of models, or if the evaluation measures a radically different criterion.<ref name="estivill" /> For example, k-means clustering can only find convex clusters, and many evaluation indexes assume convex clusters. On a data set with non-convex clusters neither the use of ''k''-means, nor of an evaluation criterion that assumes convexity, is sound.

More than a dozen of internal evaluation measures exist, usually based on the intuition that items in the same cluster should be more similar than items in different clusters.<ref name=":2">{{Citation|title=Knowledge Discovery in Databases – Part III – Clustering|date=2017|url=https://dbs.ifi.uni-heidelberg.de/files/Team/eschubert/lectures/KDDClusterAnalysis17-screen.pdf|place=[[Organization:Heidelberg University|Heidelberg University]]}}</ref>{{Rp|115–121}} For example, the following methods can be used to assess the quality of clustering algorithms based on internal criterion:

* '''[[Davies–Bouldin index]]'''
:The [[Davies–Bouldin index]] can be calculated by the following formula:
::<math>
DB = \frac {1} {n} \sum_{i=1}^{n} \max_{j\neq i}\left(\frac{\sigma_i + \sigma_j} {d(c_i,c_j)}\right)
</math>
:where ''n'' is the number of clusters, <math>c_i</math> is the [[Centroid|centroid]] of cluster <math>i</math>, <math>\sigma_i</math> is the average distance of all elements in cluster <math>i</math> to centroid <math>c_i</math>, and <math>d(c_i,c_j)</math> is the distance between centroids <math>c_i</math> and <math>c_j</math>. Since algorithms that produce clusters with low intra-cluster distances (high intra-cluster similarity) and high inter-cluster distances (low inter-cluster similarity) will have a low Davies–Bouldin index, the clustering algorithm that produces a collection of clusters with the smallest [[Davies–Bouldin index]] is considered the best algorithm based on this criterion.

*'''[[Dunn index]]'''
:The Dunn index aims to identify dense and well-separated clusters. It is defined as the ratio between the minimal inter-cluster distance to maximal intra-cluster distance. For each cluster partition, the Dunn index can be calculated by the following formula:<ref>{{Cite journal
 | last = Dunn | first = J.
 | title = Well separated clusters and optimal fuzzy partitions
 | journal = Journal of Cybernetics
 | year = 1974
 | volume = 4
 | pages = 95–104
 | doi = 10.1080/01969727408546059
}}</ref>
::<math>
D = \frac{\min_{1 \leq i < j \leq n} d(i,j)}{\max_{1 \leq k \leq n} d^{\prime}(k)} \,,
</math>
:where ''d''(''i'',''j'') represents the distance between clusters ''i'' and ''j'', and ''d'' '(''k'') measures the intra-cluster distance of cluster ''k''. The inter-cluster distance ''d''(''i'',''j'') between two clusters may be any number of distance measures, such as the distance between the centroids of the clusters. Similarly, the intra-cluster distance ''d'' '(''k'') may be measured in a variety ways, such as the maximal distance between any pair of elements in cluster&nbsp;''k''. Since internal criterion seek clusters with high intra-cluster similarity and low inter-cluster similarity, algorithms that produce clusters with high Dunn index are more desirable.
*'''[[Silhouette (clustering)|Silhouette coefficient]]'''
:The silhouette coefficient contrasts the average distance to elements in the same cluster with the average distance to elements in other clusters. Objects with a high silhouette value are considered well clustered, objects with a low value may be outliers. This index works well with ''k''-means clustering,{{citation needed|date=March 2021}} and is also used to determine the optimal number of clusters.

=== External evaluation ===
In external evaluation, clustering results are evaluated based on data that was not used for clustering, such as known class labels and external benchmarks. Such benchmarks consist of a set of pre-classified items, and these sets are often created by (expert) humans. Thus, the benchmark sets can be thought of as a [[Medicine:Gold standard (test)|gold standard]] for evaluation.<ref name="pfitzner"/> These types of evaluation methods measure how close the clustering is to the predetermined benchmark classes. However, it has recently been discussed whether this is adequate for real data, or only on synthetic data sets with a factual ground truth, since classes can contain internal structure, the attributes present may not allow separation of clusters or the classes may contain [[Anomaly detection|anomalies]].<ref name="Faerberetal2010">{{Cite conference
 | editor1-first = Xiaoli Z. | editor1-last = Fern | editor2-first = Ian | editor2-last = Davidson | editor3-first = Jennifer | editor3-last = Dy
 | book-title = MultiClust: Discovering, Summarizing, and Using Multiple Clusterings
 | title = On Using Class-Labels in Evaluation of Clusterings
 | year = 2010
 | publisher = [[Organization:Association for Computing Machinery|ACM]] [[SIGKDD]]
 | url = http://eecs.oregonstate.edu/research/multiclust/Evaluation-4.pdf
}}</ref> Additionally, from a knowledge discovery point of view, the reproduction of known knowledge may not necessarily be the intended result.<ref name="Faerberetal2010"/> In the special scenario of [[Constrained clustering|constrained clustering]], where meta information (such as class labels) is used already in the clustering process, the hold-out of information for evaluation purposes is non-trivial.<ref name="pourrajabi">{{Cite conference 
| first1 = M. | last1 = Pourrajabi
| first2 = D. | last2 = Moulavi
| first3 = R. J. G. B. | last3 = Campello
| first5 = J. | last5 = Sander
|first6 = R. |last6 = Goebel
| year = 2014
| title = Model Selection for Semi-Supervised Clustering
| book-title = Proceedings of the 17th International Conference on Extending Database Technology (EDBT)
| pages = 331–342
| doi = 10.5441/002/edbt.2014.31
}}</ref>

A number of measures are adapted from variants used to evaluate classification tasks. In place of counting the number of times a class was correctly assigned to a single data point (known as true positives), such ''pair counting'' metrics assess whether each pair of data points that is truly in the same cluster is predicted to be in the same cluster.<ref name="pfitzner"/>

As with internal evaluation, several external evaluation measures exist,<ref name=":2" />{{Rp|125–129}} for example:
* '''Purity''': Purity is a measure of the extent to which clusters contain a single class.<ref name="Christopher D. Manning, Prabhakar Raghavan & Hinrich Schutze"/> Its calculation can be thought of as follows: For each cluster, count the number of data points from the most common class in said cluster. Now take the sum over all clusters and divide by the total number of data points. Formally, given some set of clusters <math>M</math> and some set of classes <math>D</math>, both partitioning <math>N</math> data points, purity can be defined as:
::<math>
\frac{1}{N}\sum_{m\in M}\max_{d\in D}{|m \cap d|}
</math>
:This measure doesn't penalize having many clusters, and more clusters will make it easier to produce a high purity. A purity score of 1 is always possible by putting each data point in its own cluster. Also, purity doesn't work well for imbalanced data, where even poorly performing clustering algorithms will give a high purity value. For example, if a size 1000 dataset consists of two classes, one containing 999 points and the other containing 1 point, then every possible partition will have a purity of at least 99.9%. 
* '''[[Rand index]]'''<ref>{{Cite journal
 | first = W. M. | last = Rand
 | title = Objective criteria for the evaluation of clustering methods
 | journal = [[Organization:Journal of the American Statistical Association|Journal of the American Statistical Association]]
 | volume = 66
 | pages = 846–850
 | year = 1971
 | doi = 10.2307/2284239
 | issue = 336
 | publisher = American Statistical Association
 | jstor = 2284239
| arxiv = 1704.01036
 }}</ref>
:The Rand index computes how similar the clusters (returned by the clustering algorithm) are to the benchmark classifications. It can be computed using the following formula:
::<math>
RI = \frac {TP + TN} {TP + FP + FN + TN}
</math>
:where  <math>TP</math> is the number of true positives, <math>TN</math> is the number of true negatives, <math>FP</math> is the number of false positives, and <math>FN</math> is the number of false negatives. The instances being counted here are the number of correct ''pairwise'' assignments. That is, <math>TP</math> is the number of pairs of points that are clustered together in the predicted partition and in the ground truth partition, <math>FP</math> is the number of pairs of points that are clustered together in the predicted partition but not in the ground truth partition etc. If the dataset is of size N, then <math>TP + TN + FP + FN = \binom{N}{2}</math>. 
One issue with the [[Rand index]] is that false positives and false negatives are equally weighted. This may be an undesirable characteristic for some clustering applications. The F-measure addresses this concern,{{Citation needed|date=May 2018|reason=it does not achieve the same for chance correction as ARI}} as does the chance-corrected adjusted Rand index.
*'''F-measure'''
:The F-measure can be used to balance the contribution of false negatives by weighting recall through a parameter <math>\beta \geq 0</math>. Let '''precision''' and '''recall''' (both external evaluation measures in themselves) be defined as follows:
::<math>
P = \frac {TP } {TP + FP }
</math>
::<math>
R = \frac {TP } {TP + FN}
</math>
:where <math>P</math> is the precision rate and <math>R</math> is the recall rate. We can calculate the F-measure by using the following formula:<ref name="Christopher D. Manning, Prabhakar Raghavan & Hinrich Schutze"/>
::<math>
F_{\beta} = \frac {(\beta^2 + 1)\cdot P \cdot R } {\beta^2 \cdot P + R}
</math>
:When <math>\beta=0</math>, <math>F_{0}=P</math>. In other words, recall has no impact on the F-measure when <math>\beta=0</math>, and increasing <math>\beta</math> allocates an increasing amount of weight to recall in the final F-measure.
:Also <math>TN</math> is not taken into account and can vary from 0 upward without bound.
* '''Jaccard index'''
:The Jaccard index is used to quantify the similarity between two datasets. The Jaccard index takes on a value between 0 and 1. An index of 1 means that the two dataset are identical, and an index of 0 indicates that the datasets have no common elements. The Jaccard index is defined by the following formula:
::<math>
J(A,B) = \frac {|A \cap B| } {|A \cup B|} = \frac{TP}{TP + FP + FN}
</math>
:This is simply the number of unique elements common to both sets divided by the total number of unique elements in both sets.
:Note that <math>TN</math> is not taken into account.
* '''[[Sørensen–Dice coefficient|Dice index]]'''
:The Dice symmetric measure doubles the weight on <math>TP</math> while still ignoring <math>TN</math>:
::<math>
DSC = \frac{2TP}{2TP + FP + FN}
</math>
* '''Fowlkes–Mallows index'''<ref>{{cite journal | last1 = Fowlkes | first1 = E. B. | last2 = Mallows | first2 = C. L. | year = 1983 | title = A Method for Comparing Two Hierarchical Clusterings | jstor = 2288117 | journal = Journal of the American Statistical Association | volume = 78 | issue = 383| pages = 553–569 | doi = 10.1080/01621459.1983.10478008 }}</ref>
:The Fowlkes–Mallows index computes the similarity between the clusters returned by the clustering algorithm and the benchmark classifications. The higher the value of the Fowlkes–Mallows index the more similar the clusters and the benchmark classifications are. It can be computed using the following formula:
::<math>
FM = \sqrt{ \frac {TP}{TP+FP} \cdot \frac{TP}{TP+FN}  }
</math>
:where  <math>TP</math> is the number of true positives, <math>FP</math> is the number of false positives, and <math>FN</math> is the number of false negatives. The <math>FM</math> index is the geometric mean of the precision and recall <math>P</math> and <math>R</math>, and is thus also known as the G-measure, while the F-measure is their harmonic mean.<ref name="powers">{{cite conference | last = Powers | first = David | date = 2003 | conference = International Conference on Cognitive Science | pages = 529–534 | title = Recall and Precision versus the Bookmaker }}</ref><ref>{{cite journal | last1 = Arabie | first1 = P. | year = 1985| title = Comparing partitions | journal = Journal of Classification | volume = 2 | issue = 1| page = 1985 | doi = 10.1007/BF01908075 | s2cid = 189915041 }}</ref> Moreover, precision and recall are also known as Wallace's indices <math>B^I</math> and <math>B^{II}</math>.<ref>{{cite journal | last1 = Wallace | first1 = D. L. | year = 1983 | title = Comment | journal = Journal of the American Statistical Association | volume = 78 | issue = 383| pages = 569–579 | doi=10.1080/01621459.1983.10478009}}</ref> Chance normalized versions of recall, precision and G-measure correspond to Informedness, [[Social:Markedness|Markedness]] and [[Biology:Matthews correlation coefficient|Matthews Correlation]] and relate strongly to [[Cohen's kappa|Kappa]].<ref name="kappa">{{cite conference | last1 = Powers | first1 = David | title = The Problem with Kappa | conference = European Chapter of the Association for Computational Linguistics | date = 2012 | pages = 345–355}}</ref>

* '''Chi Index'''<ref>{{Cite journal |last1=Luna-Romera |first1=José María |last2=Martínez-Ballesteros |first2=María |last3=García-Gutiérrez |first3=Jorge |last4=Riquelme |first4=José C. |date=June 2019 |title=External clustering validity index based on chi-squared statistical test |url=https://linkinghub.elsevier.com/retrieve/pii/S0020025519301550 |journal=Information Sciences |language=en |volume=487 |pages=1–17 |doi=10.1016/j.ins.2019.02.046|hdl=11441/132081 |s2cid=93003939 |hdl-access=free }}</ref> is a external validation index that measure the clustering results by applying the [[Chi-squared test|chi-squared statistic]]. This index scores positively the fact that the labels are as sparse as possible across the clusters, i.e., that each cluster has as few different labels as possible. The higher the value of the Chi Index the greater the relationship between the resulting clusters and the label used.
* The '''[[Mutual information|mutual information]]''' is an [[Information theory|information theoretic]] measure of how much information is shared between a clustering and a ground-truth classification that can detect a non-linear similarity between two clusterings. [[Adjusted mutual information|Normalized mutual information]] is a family of corrected-for-chance variants of this that has a reduced bias for varying cluster numbers.<ref name="pfitzner" />
* '''[[Confusion matrix]]'''
:A confusion matrix can be used to quickly visualize the results of a classification (or clustering) algorithm. It shows how different a cluster is from the gold standard cluster.

=== Cluster tendency ===

To measure cluster tendency is to measure to what degree clusters exist in the data to be clustered, and may be performed as an initial test, before attempting clustering. One way to do this is to compare the data against random data. On average, random data should not have clusters.

*'''[[Hopkins statistic]]'''
:There are multiple formulations of the [[Hopkins statistic]].<ref>{{Cite journal
  | title = A new method for determining the type of distribution of plant individuals
  | last1 = Hopkins | first1 = Brian
  | last2 = Skellam | first2 = John Gordon
  | journal = Annals of Botany
  | volume =18
  | number = 2
  | pages = 213–227
  | year = 1954
  | publisher = Annals Botany Co
  | doi=10.1093/oxfordjournals.aob.a083391
}}</ref> A typical one is as follows.<ref>{{Cite book
 | last = Banerjee | first = A.
 | title = 2004 IEEE International Conference on Fuzzy Systems (IEEE Cat. No.04CH37542)
 | chapter = Validating clusters using the Hopkins statistic
 | s2cid = 36701919
 | volume = 1
 | pages = 149–153
 | doi = 10.1109/FUZZY.2004.1375706
| year = 2004
 | isbn = 978-0-7803-8353-1
 }}</ref> Let <math>X</math> be the set of <math>n</math> data points in <math>d</math> dimensional space. Consider a random sample (without replacement) of <math>m \ll n</math> data points with members <math>x_i</math>. Also generate a set <math>Y</math> of <math>m</math> uniformly randomly distributed data points. Now define two distance measures, <math>u_i</math> to be the distance of <math>y_i \in Y</math> from its nearest neighbor in X and <math>w_i</math> to be the distance of <math>x_i \in X</math> from its nearest neighbor in X. We then define the Hopkins statistic as:
:: <math>
H=\frac{\sum_{i=1}^m{u_i^d}}{\sum_{i=1}^m{u_i^d}+\sum_{i=1}^m{w_i^d}} \,,
</math>
:With this definition, uniform random data should tend to have values near to 0.5, and clustered data should tend to have values nearer to 1.
:However, data containing just a single Gaussian will also score close to 1, as this statistic measures deviation from a ''uniform'' distribution, not [[Multimodal distribution|multimodality]], making this statistic largely useless in application (as real data never is remotely uniform).

== Applications ==

===Biology, computational biology and bioinformatics===


; [[Biology:Plant|Plant]] and [[Biology:Animal|animal]] [[Earth:Ecology|ecology]]
:Cluster analysis is used to describe and to make spatial and temporal comparisons of communities (assemblages) of organisms in heterogeneous environments. It is also used in [[Biology:Systematics|plant systematics]] to generate artificial [[Biology:Phylogeny|phylogenies]] or clusters of organisms (individuals) at the species, genus or higher level that share a number of attributes.
; Transcriptomics
:Clustering is used to build groups of genes with related expression patterns (also known as coexpressed genes) as in [[HCS clustering algorithm]].<ref>{{Cite journal|last=Johnson|first=Stephen C.|s2cid=930698|date=1967-09-01|title=Hierarchical clustering schemes|journal=Psychometrika|language=en|volume=32|issue=3|pages=241–254|doi=10.1007/BF02289588|pmid=5234703|issn=1860-0980}}</ref><ref>{{Cite journal|last1=Hartuv|first1=Erez|last2=Shamir|first2=Ron|date=2000-12-31|title=A clustering algorithm based on graph connectivity|journal=Information Processing Letters|volume=76|issue=4|pages=175–181|doi=10.1016/S0020-0190(00)00142-3|issn=0020-0190}}</ref> Often such groups contain functionally related proteins, such as [[Biology:Enzyme|enzyme]]s for a specific [[Biology:Metabolic pathway|pathway]], or genes that are co-regulated. High throughput experiments using [[Biology:Expressed sequence tag|expressed sequence tag]]s (ESTs) or [[Biology:DNA microarray|DNA microarray]]s can be a powerful tool for genome annotation{{snd}}a general aspect of [[Biology:Genomics|genomics]].
; [[Biology:Sequence analysis|Sequence analysis]]
:[[Biology:Sequence clustering|Sequence clustering]] is used to group homologous sequences into [[Biology:List of gene families|gene families]].<ref>{{Cite journal|last1=Remm|first1=Maido|last2=Storm|first2=Christian E. V.|last3=Sonnhammer|first3=Erik L. L.|date=2001-12-14|title=Automatic clustering of orthologs and in-paralogs from pairwise species comparisons11Edited by F. Cohen|journal=Journal of Molecular Biology|volume=314|issue=5|pages=1041–1052|doi=10.1006/jmbi.2000.5197|issn=0022-2836|pmid=11743721}}</ref> This is a very important concept in [[Biology:Bioinformatics|bioinformatics]], and [[Biology:Evolutionary biology|evolutionary biology]] in general. See evolution by [[Biology:Gene duplication|gene duplication]].
; High-throughput [[Biology:Genotype|genotyping]] platforms
:Clustering algorithms are used to automatically assign genotypes.<ref>{{Cite journal|last1=Botstein|first1=David|last2=Cox|first2=David R.|last3=Risch|first3=Neil|last4=Olshen|first4=Richard|last5=Curb|first5=David|last6=Dzau|first6=Victor J.|last7=Chen|first7=Yii-Der I.|last8=Hebert|first8=Joan|last9=Pesich|first9=Robert|date=2001-07-01|title=High-Throughput Genotyping with Single Nucleotide Polymorphisms|url=http://genome.cshlp.org/content/11/7/1262|journal=Genome Research|language=en|volume=11|issue=7|pages=1262–1268|doi=10.1101/gr.157801|issn=1088-9051|pmid=11435409|pmc=311112}}</ref>
; [[Human genetic clustering]]
:The similarity of genetic data is used in clustering to infer population structures.

===[[Medicine:Medicine|Medicine]]===
; [[Medicine:Medical imaging|Medical imaging]]
:On PET scans, cluster analysis can be used to differentiate between different types of [[Biology:Tissue|tissue]] in a three-dimensional image for many different purposes.<ref>{{cite journal |last1=Filipovych |first1=Roman |last2=Resnick |first2=Susan M. |last3=Davatzikos |first3=Christos|title=Semi-supervised Cluster Analysis of Imaging Data |journal=NeuroImage |date=2011|volume=54 |issue=3 |pages=2185–2197 |doi=10.1016/j.neuroimage.2010.09.074|pmc=3008313 |pmid=20933091}}</ref>
; Analysis of antimicrobial activity
:Cluster analysis can be used to analyse patterns of antibiotic resistance, to classify antimicrobial compounds according to their mechanism of action, to classify antibiotics according to their antibacterial activity.
; IMRT segmentation
:Clustering can be used to divide a fluence map into distinct regions for conversion into deliverable fields in MLC-based Radiation Therapy.

===Business and marketing===
; [[Finance:Market research|Market research]]
:Cluster analysis is widely used in market research when working with multivariate data from surveys and test panels. Market researchers use cluster analysis to partition the general [[Social:Population|population]] of consumers into market segments and to better understand the relationships between different groups of consumers/potential customers, and for use in [[Finance:Market segmentation|market segmentation]], [[Social:Positioning (marketing)|product positioning]], [[Engineering:New product development|new product development]] and selecting test markets.
; Grouping of shopping items
:Clustering can be used to group all the shopping items available on the web into a set of unique products. For example, all the items on eBay can be grouped into unique products (eBay does not have the concept of a SKU).

===[[World Wide Web]]===
; Social network analysis
:In the study of [[Social:Social network|social network]]s, clustering may be used to recognize communities within large groups of people.
; Search result grouping
:In the process of intelligent grouping of the files and websites, clustering may be used to create a more relevant set of search results compared to normal search engines like [[Company:Google|Google]]{{citation needed|date=July 2018}}. There are currently a number of web-based clustering tools such as Clusty. It also may be used to return a more comprehensive set of results in cases where a search term could refer to vastly different things. Each distinct use of the term corresponds to a unique cluster of results, allowing a ranking algorithm to return comprehensive results by picking the top result from each cluster.<ref name="mitpressjournals.org">{{Cite journal |doi = 10.1162/COLI_a_00148|title = Clustering and Diversifying Web Search Results with Graph-Based Word Sense Induction|journal = Computational Linguistics|volume = 39|issue = 3|pages = 709–754|year = 2013|last1 = Di Marco|first1 = Antonio|last2 = Navigli|first2 = Roberto|s2cid = 1775181}}</ref>
; Slippy map optimization
:[[Social:Flickr|Flickr]]'s map of photos and other map sites use clustering to reduce the number of markers on a map.{{cn|date=May 2023}} This makes it both faster and reduces the amount of visual clutter.

===[[Computer science]]===
; [[Software evolution]]
:Clustering is useful in software evolution as it helps to reduce legacy properties in code by reforming functionality that has become dispersed. It is a form of restructuring and hence is a way of direct preventative maintenance.
; [[Image segmentation]]
:Clustering can be used to divide a [[Digital data|digital]] [[Image|image]] into distinct regions for border detection or object recognition.<ref name="panSearch">Bewley, A., & Upcroft, B. (2013). Advantages of Exploiting Projection Structure for Segmenting Dense 3D Point Clouds. In Australian Conference on Robotics and Automation [http://www.araa.asn.au/acra/acra2013/papers/pap148s1-file1.pdf]</ref>
; Evolutionary algorithms
:Clustering may be used to identify different niches within the population of an evolutionary algorithm so that reproductive opportunity can be distributed more evenly amongst the evolving species or subspecies.
; Recommender systems
:Recommender systems are designed to recommend new items based on a user's tastes.  They sometimes use clustering algorithms to predict a user's preferences based on the preferences of other users in the user's cluster.
; [[Markov chain Monte Carlo|Markov chain Monte Carlo methods]]
:Clustering is often utilized to locate and characterize extrema in the target distribution.
; [[Anomaly detection]]
:Anomalies/outliers are typically – be it explicitly or implicitly – defined with respect to clustering structure in data.
; [[Natural language processing]]
:Clustering can be used to resolve lexical ambiguity.<ref name="mitpressjournals.org"/>
; [[DevOps]]
:Clustering has been used to analyse the effectiveness of DevOps teams.<ref name="stateofdevopsreport">{{cite journal |date=29 September 2022 |title=2022 Accelerate State of DevOps Report |journal= |pages=8, 14, 74}} [https://services.google.com/fh/files/misc/2022_state_of_devops_report.pdf]</ref>

===Social science===
; [[Sequence analysis in social sciences]]
:Cluster analysis is used to identify patterns of family life trajectories, professional careers, and daily or weekly time use for example. 
; Crime analysis
:Cluster analysis can be used to identify areas where there are greater incidences of particular types of crime. By identifying these distinct areas or "hot spots" where a similar crime has happened over a period of time, it is possible to manage law enforcement resources more effectively.
; [[Educational data mining]]
:Cluster analysis is for example used to identify groups of schools or students with similar properties.
; Typologies
:From poll data, projects such as those undertaken by the Pew Research Center use cluster analysis to discern typologies of opinions, habits, and demographics that may be useful in politics and marketing.

===Others===

; Field robotics
:Clustering algorithms are used for robotic situational awareness to track objects and detect outliers in sensor data.<ref>{{cite journal | last1 = Bewley | first1 = A. | title = Real-time volume estimation of a dragline payload | journal = IEEE International Conference on Robotics and Automation | volume = 2011 | pages = 1571–1576 |display-authors=0}}</ref>

; [[Chemistry:Mathematical chemistry|Mathematical chemistry]]
:To find structural similarity, etc., for example, 3000 chemical compounds were clustered in the space of 90 [[Chemistry:Topological index|topological indices]].<ref>{{cite journal | last1 = Basak | first1 = S.C. | last2 = Magnuson | first2 = V.R. | last3 = Niemi | first3 = C.J. | last4 = Regal | first4 = R.R. | title = Determining Structural Similarity of Chemicals Using Graph Theoretic Indices | journal = Discr. Appl. Math. |volume=19 | issue = 1–3 | year= 1988 | pages = 17–44 | doi=10.1016/0166-218x(88)90004-2| doi-access = free }}</ref>

; [[Earth:Climatology|Climatology]]
:To find weather regimes or preferred sea level pressure atmospheric patterns.<ref>{{cite journal | last1 = Huth | first1 = R. | year = 2008 | title = Classifications of Atmospheric Circulation Patterns: Recent Advances and Applications | journal = Ann. N.Y. Acad. Sci. | volume = 1146 | issue = 1 | pages = 105–152 |display-authors=0| bibcode = 2008NYASA1146..105H | doi = 10.1196/annals.1446.019 | pmid = 19076414 | s2cid = 22655306 | url = https://opus.bibliothek.uni-augsburg.de/opus4/files/40082/40082.pdf }}</ref>

; Finance
:Cluster analysis has been used to cluster stocks into sectors.<ref>{{Cite journal|last=Arnott|first=Robert D.|date=1980-11-01|title=Cluster Analysis and Stock Price Comovement|journal=Financial Analysts Journal|volume=36|issue=6|pages=56–62|doi=10.2469/faj.v36.n6.56|issn=0015-198X}}</ref>

;Petroleum geology
:Cluster analysis is used to reconstruct missing bottom hole core data or missing log curves in order to evaluate reservoir properties.

; Geochemistry
:The clustering of chemical properties in different sample locations.

== See also ==

=== Specialized types of cluster analysis ===
* [[Automatic clustering algorithms]]
* [[Balanced clustering]]
* [[Clustering high-dimensional data]]
* [[Conceptual clustering]]
* [[Consensus clustering]]
* [[Constrained clustering]]
* [[Community structure#Algorithms for finding communities|Community detection]]
* [[Data stream clustering]]
* [[HCS clustering algorithm|HCS clustering]]
* [[Biology:Sequence clustering|Sequence clustering]]
* [[Spectral clustering]]

=== Techniques used in cluster analysis ===
* [[Artificial neural network]] (ANN)
* [[Nearest neighbor search]]
* [[Neighbourhood components analysis]]
* [[Latent class model|Latent class analysis]]
* [[Affinity propagation]]

=== Data projection and preprocessing ===
* Dimension reduction
* [[Principal component analysis]]
* [[Multidimensional scaling]]

=== Other ===
* [[Cluster-weighted modeling]]
* [[Curse of dimensionality]]
* Determining the number of clusters in a data set
* [[Parallel coordinates]]
* [[Structured data analysis (statistics)|Structured data analysis]]
* [[Linear separability]]

== References ==
{{Reflist}}

{{Clear}}
{{Differentiable computing}}
{{Statistics|analysis}}

{{DEFAULTSORT:Cluster Analysis}}
[[Category:Cluster analysis| ]]
[[Category:Data mining]]
[[Category:Geostatistics]]

{{Sourceattribution|Cluster analysis}}