K-Means Clustering

K-means clustering is an unsupervised method that partitions observations into a chosen number of clusters by iteratively assigning each observation to the nearest cluster mean and updating those means.

K-평균 군집분석은 각 관측치를 가장 가까운 군집 평균에 할당하고 그 평균을 갱신하는 과정을 반복하여 관측치를 정해진 수의 군집으로 나누는 비지도 학습 방법입니다.

How do I group observations into segments based on several numeric variables?

여러 수치형 변수를 기준으로 관측치를 세분집단으로 어떻게 나누나요?

K-means clustering divides observations into k groups so that each observation belongs to the cluster with the nearest centroid, minimizing within-cluster variance. It fits continuous, standardized variables when clusters are expected to be roughly spherical and of similar size, with k chosen using the elbow method or silhouette scores.

When to use K-Means Clustering

• Customer or respondent segmentation on several continuous variables
• Large datasets where a fast, simple partition is needed
• Exploratory grouping when roughly compact, similar-sized clusters are expected

When to use something else

• Irregularly shaped clusters or many outliers: use DBSCAN
• Probabilistic cluster membership is needed: use a Gaussian mixture model

Assumptions and requirements

• Continuous variables, standardized to comparable scales
• Number of clusters specified in advance
• Roughly spherical clusters of similar size
• Multiple random starts to avoid poor local solutions

What the analysis reports

• Cluster assignment for each observation
• Cluster centroids (profiles)
• Within-cluster sum of squares and elbow plot
• Average silhouette width
• Cluster sizes

How to report it

K-means clustering identified [k] clusters, selected using the elbow method and silhouette analysis (average silhouette width = [value]).

Related methods

• Hierarchical Clustering
• Gaussian Mixture Model (GMM)
• DBSCAN

Run K-Means Clustering on Quorant