Clustering analysis: selecting the right clustering algorithm

Practicing Machine Learning Interview Questions in Python

Lisa Stuart

Data Scientist

Clustering algorithms

Features >> Observations
Model training more challenging
Rely on distance calculations
Most commonly used unsupervised technique

Practical applications of clustering

Customer segmentation
Document classification
Insurance/transaction fraud detection
Image segmentation
Anomaly detection
Many more...

Distance metrics: Manhattan (taxicab) distance

Manhattan plot

¹ https://en.wikipedia.org/wiki/Taxicab_geometry

Distance metrics: Euclidian distance

Euclidian plot

¹ http://rosalind.info/glossary/euclidean-distance/

K-means

K-means steps

Initial centroids
Assign each observation to nearest centroid
Create new centroids
Repeat steps 2 and 3

¹ http://sherrytowers.com/2013/10/24/k-means-clustering/

Hierarchical agglomerative clustering

Hierarchical clustering dendrogram

¹ https://www.datanovia.com/en/lessons/agglomerative-hierarchical-clustering/

Agglomerative clustering linkage

Ward linkage
Maximum/complete linkage
Average linkage
Single linkage

Selecting a clustering algorithm

Cluster distances

Cluster stability assessment
K-means and HC use Euclidian distance
Inter- and intra-cluster distances

"An appropriate dissimilarity measure is far more important in obtaining success with clustering than choice of clustering algorithm." - from Elements of Statistical Learning

¹ https://slideplayer.com/slide/8363774/

Clustering functions

Function/method	returns
`sklearn.cluster.Kmeans`	K-Means clustering algorithm
`sklearn.cluster.AgglomerativeClustering`	Agglomerative clustering algorithm
`kmeans.inertia_`	SS distances of observations to closest cluster center
`scipy.cluster.hierarchy` as `sch`	Hierachical clustering for dendrograms
`sch.dendrogram()`	Dendrogram function

Let's practice!

Practicing Machine Learning Interview Questions in Python