Distance Measures

EUCLIDIAN DISTANCE:

This is the most commonly used distance measure. It is a special case of Minkowski's distance when the value of p = 2. In layman's term Euclidian distance can be defined as the shortest distance between the points.

USE CASES:

Works well with low dimensional data
Good compatibility with KNN and HDBSCAN

cons:

Cannot be used for high dimensional data
Need to normalize before using

Manhattan DISTANCE:

Not all the cases, shortest/ diagonal distance can be computed or doesn't meet the requirements. Thats where Manhattan distance comes in. It is also a special case of Minkowski's distance when the value of p = 1. Manhattan distance is defined as the distance between the real valued vectors ie the actual distance without considering the diagonal movements. It is also referred to as taxi cab or city block distance.

USE CASES:

Works well with discrete/ binary attributes

cons:

Magnitude is high as the shortest distance is neglected
Less intutive for high dimension data

COSINE SIMILARITY:

This is the next most commonly used distance measure after Euclidian. It is computed mainly to counter the high dimensional data problem faced by Euclidian distance. It is computed by measuring the cosine angle between the vectors.

USE CASES:

Works well with high dimensional data where magnitude is not a major factor
Text Analysis

cons:

Magnitude is neglected
Considers only direction

MINKOWSKI DISTANCE:

Minkowski distance is used in normed vector space. There are three major requirements for this distance calculation.

Vector with length zero
Scalar Factor
Triangle Inequality

Overview

Partitional Clustering

Hierarchical Clustering

Density Based Clustering

Page updated

Google Sites

Report abuse