KMeans

E97072

clustering algorithm iterative optimization algorithm partition-based clustering method unsupervised learning algorithm

KMeans is a popular unsupervised machine learning algorithm used for partitioning data into a specified number of clusters based on feature similarity.

Statements (50)

Predicate	Object
instanceOf	clustering algorithm → iterative optimization algorithm → partition-based clustering method → unsupervised learning algorithm →
advantage	computationally efficient for large datasets → scales linearly with number of samples and clusters in practice → simple to implement →
alsoKnownAs	Lloyd’s algorithm → k-means clustering →
assumes	Euclidean feature space in standard form → clusters are roughly spherical → clusters have similar size →
basedOn	minimization of within-cluster sum of squares →
canUseDistanceMetric	other Lp distances with modifications →
commonlyUsedIn	customer segmentation → document clustering → image compression → pattern recognition →
convergesWhen	change in objective function is below a threshold → cluster assignments no longer change →
distanceMetric	Euclidean distance (standard) →
implementedIn	Apache Spark MLlib → MATLAB Statistics and Machine Learning Toolbox → R stats and cluster packages → scikit-learn →
input	number of clusters k → set of data points →
limitation	cannot automatically determine optimal number of clusters → may converge to local minima → not robust to noise and outliers → performs poorly on non-spherical clusters →
objectiveFunction	minimize sum of squared distances between points and their assigned cluster centroid →
optimizationProblem	NP-hard in general →
relatedAlgorithm	Gaussian mixture models → fuzzy c-means → k-medoids →
requires	numerical feature representation → predefined number of clusters k →
sensitiveTo	feature scaling → initialization → outliers →
step	assign each point to nearest centroid → iterate assignment and update until convergence → recompute centroids as mean of assigned points →
typicalInitialization	k-means++ initialization → random selection of initial centroids →
usedFor	data compression → partitioning data into k clusters → prototype-based clustering → vector quantization →

Referenced by (1)

Subject (surface form when different)	Predicate
scikit-learn →	hasConcept