Multidimensional Scaling
Multidimensional Scaling (MDS) is a technique used in data analysis and visualization to
represent high-dimensional data in a lower-dimensional space, usually two or three dimensions.
The primary goal of MDS is to capture the relative distances or dissimilarities between data
points, making it easier to visualize patterns or groupings that may not be apparent in higher
dimensions.
Key Concepts in MDS:
1. Dissimilarity Matrix: MDS with a matrix representing the pairwise distances or
dissimilarities between items. These distances can be calculated using various
measures, such as Euclidean distance for continuous data or more specialized measures
for categorical data.
2. Dimensional Reduction: MDS transforms the data into a specified number of
dimensions (e.g., 2D or 3D) while attempting to maintain the original dissimilarities as
accurately as possible in the new space. The goal is to find a layout where the distances
between points in the reduced space are close to the original distances in the high-
dimensional space.
3. Stress Function: MDS algorithms typically minimize a "stress" function, which measures
the difference between the distances in the original space and those in the reduced
space. Lower stress values indicate a more accurate representation of the original
distances in the new dimensions
Types of MDS:
• Classical MDS: Assumes linear relationships and is typically used for Euclidean
distance-based data. It finds a coordinate matrix that preserves distances in a
lower-dimensional space.
• Non-metric MDS: Useful for data with rank-based distances, where only the
relative order of distances matters. This version of MDS is more flexible and can
be used for a wider range of distance measures.
MDS is especially useful for visualizing and interpreting complex datasets, revealing underlying
structures and relationships by presenting them in an interpretable, lower-dimensional format.
Example
Given Dissimilarity Matrix D
A B C D
A 0 2 5 6
B 2 0 3 5
C 5 3 0 1
D 6 5 1 0
