Clustering

PCA and Clustering similarities

both trying to simplify the data

classification

group data into specific categories using labeled dataset

clustering

group similar data instances together using unlabeled dataset, clusters are not necessarily aligned with classification

PCA vs clustering

pca maximizes the variance of PC scores

clustering

minimizes within cluster variance

PCA

maximize the variance of PC scores

Euclidean distance

used for continuous data

mahalanobis

adjusts for variable correlations

k means clustering

minimize within cluster variation

centroid

the vector the p feature means for the observation

k mediods

choose data points as the centers (mediods)

mediods

most centrally located point in the cluster

k modes clustering optimal k

have a scree plot of dissimilarities and find the elbow point.

k-modes procedure

count dissimilarities

k- prototype

designed to handle mixed datasets (both numerical and categorical) combines k-means and k-modes

Silhouette

measures how similar an object is to its own cluster compared to other clusters.

silhouette range

-1 to +1