Source-linked AI summary
Semi-supervised clustering methods
Eric Bair
TL;DR
Semi-supervised clustering uses partial labels, pairwise constraints, or outcomes to guide clustering when conventional unsupervised methods lack this information. This review surveys these methods, emphasizing k-means modifications, and reports applications including improved constraint selection and biologically relevant clusters.
Problem
Semi-supervised clustering addresses settings where some cluster assignments, pairwise relationships, or related outcomes are known but observations remain unlabeled.
Method
The review surveys semi-supervised clustering algorithms, primarily modifications of k-means, including constrained, outcome-associated, and sparse clustering methods.
Results
Active PCKmeans outperforms generic PCKmeans, while outcome-guided methods identify biologically relevant clusters and perform well in high-dimensional data.
Takeaways & Limitations
Semi-supervised clustering provides multiple ways to incorporate partial labels, constraints, or outcomes into cluster analysis across application settings.
Takeaways & Limitations
The review notes limited comparative evidence on constrained-clustering algorithms and limited research using modern high-throughput genetic data beyond DNA microarrays.
Abstract
from arXiv · showhide
Cluster analysis methods seek to partition a data set into homogeneous subgroups. It is useful in a wide variety of applications, including document processing and modern genetics. Conventional clustering methods are unsupervised, meaning that there is no outcome variable nor is anything known about the relationship between the observations in the data set. In many situations, however, information about the clusters is available in addition to the values of the features. For example, the cluster labels of some observations may be known, or certain observations may be known to belong to the same cluster. In other cases, one may wish to identify clusters that are associated with a particular outcome variable. This review describes several clustering algorithms (known as "semi-supervised clustering" methods) that can be applied in these situations. The majority of these methods are modifications of the popular k-means clustering method, and several of them will be described in detail. A brief description of some other semi-supervised clustering algorithms is also provided.
Traditional (Unsupervised) Clustering Methods
Traditional clustering methods include k-means and hierarchical clustering, with k-means assigning quantitative observations to clusters by minimizing squared Euclidean within-cluster variation. K-means converges but may reach a local minimum, and it requires selecting the number of clusters.
- K-means clustering: K-means is a popular method for quantitative data that measures distances between observations using squared Euclidean distance.
- K-means clustering: The k-means algorithm assigns each observation to a cluster to minimize the within-cluster sum of squares, or WCSS.
- K-means clustering: K-means is guaranteed to converge but may reach a local minimum, so it should be repeated with different initial clusters and the minimum-WCSS solution selected.
- K-means clustering: Choosing K is necessary because Wk always decreases as k increases, so K cannot be selected simply by minimizing Wk.
- Limitations of traditional clustering: Because conventional clustering methods do not use outcome values, they may identify clusters unrelated to the outcome instead of the clusters of interest.
Cross-References
The paper relates its discussion to cluster analysis, K-means clustering, and large datasets.
- Related topics include cluster analysis, K-means clustering, and large datasets.