Compare and rank

Andrews curves

Maps each observation's numeric variables to coefficients of a Fourier-like curve.

Plot
Multivariate comparison
Clustering

What it answers

Which multivariate profiles resemble each other?

Shape: Curves built from variable coefficients
What it shows: Similar profiles and unusual records.

Example

The question

Do our customer segments have distinct overall profiles?

Andrews curves for 15 customers in three segments

Each customer becomes one curve built from four standardised measures. Curves of the same colour bunch together, so the three segments have distinct profiles.

Illustrative data. It belongs to no company.

When to use it

Use it when

Exploratory clustering of standardised numeric profiles.

Avoid it when

Unscaled mixed units or an audience needing direct variable lookup.

How to read it

Compare curve shapes after checking variable order and scaling.

Note

Variable scaling and order change the curves.

The data you need

Layout: Wide observation table
Fields: ObservationID: text; V1...Vp: numeric columns
One row is: One observation with several measured numeric variables. What is the grain?
Rules: Align variables by observation. Decide scaling, missing-data handling and which columns are features.
Optional: Optional Group. PCA, clustering and correlations require calculations before rendering their derived results.
Preparation: Order and scale variables, then transform each observation into an Andrews function using a documented basis.
Units: State the unit of every coordinate/variable. A unitless transformed or standardised scale must be identified; no common denominator is assumed.

A wide table: one row per item, with a separate column for each measure or period.

Worked example

Two identical feature vectors create identical Andrews curves. Changing the third variable alters one curve's shape; closeness depends on preprocessing.

ID	V1	V2	V3
A	10	30	5
B	12	20	8
C	18	25	6

All charts in the library