ML0059 K-means II

K-Means clustering is a widely used unsupervised learning algorithm that partitions data points into K clusters, where each point belongs to the cluster with the nearest mean. While it is computationally efficient and easy to implement, it relies on prior specification of the number of clusters, assumes spherical clusters, and is sensitive to initialization and outliers.

Objective Function of K-means:

Where:
is the number of clusters.
is the set of points in cluster .
is the centroid of cluster .
is a data point assigned to cluster .

Main Benefits of K-means:
(1) Simple & Efficient: Fast to compute, easy to implement.
(2) Scalable: Handles large datasets well.
(3) Unsupervised Learning: Requires no labeled data.
(4) Interpretable: Cluster centroids are intuitive and interpretable.
(5) Works Well on Spherical Clusters: Performs best when clusters are compact and well-separated.

Main Disadvantages of K-means:
(1) Must Specify : The number of clusters must be known in advance.
(2) Sensitive to Initialization: Poor starting points may lead to suboptimal clustering.
(3) Assumes Spherical Clusters: Fails on clusters with irregular shapes or varying densities.
(4) Affected by Outliers: Outliers can skew centroids and degrade performance.
(5) Only Uses Euclidean Distance: Not suitable for non-numeric or categorical features.

Examples below demonstrate K-Means performance on spherical clusters and Irregular shape clusters.