Mahalanobis distance

As explained in Outlier Detection in Python

mahalanobis() is an estimator method that computes Mahalanobis distances for observations.

Mahalanobis distance measures how far an observation lies from a data distribution while accounting for its center, feature variances, and covariances. Unlike Euclidean distance, it reflects the distribution’s shape: movement along a high-variance direction counts less than equally sized movement along a low-variance direction. The method is therefore useful for assessing outlierness in correlated tabular data. section 5.5 section 5.5.1

Figure 5.6 The middle pane is an example of using Euclidean distance from the center of a dataset to measure the outlierness of datapoints. Here, point A is a stronger outlier than point B but is the same distance from the center using Euclidean distance. The right pane is an example using distances based on fitting an ellipse to the data and provides more meaningful distance measures, allowing us to consider A farther from the center than B.

Compare Euclidean and ellipse-adjusted distances: points equally distant under Euclidean geometry can receive different Mahalanobis distances when the data shape makes one more unusual. section 5.5

The calculation incorporates the inverse covariance matrix, effectively transforming features into an uncorrelated, equal-variance space before applying Euclidean distance. This interpretation connects the method to principal component analysis. Standard use assumes one cluster with roughly Gaussian, or at least unimodal, dimensions. Those assumptions can fail, so flagged observations require investigation rather than automatic acceptance as meaningful anomalies. section 5.5.1

A practical implementation proceeds by:

• Fitting a covariance estimator to the dataset and using its estimated center and covariance matrix. section 5.5.1
• Preferring robust estimators such as minimum covariance determinant or minimum volume ellipsoid when outliers could distort covariance estimates through masking or swamping. section 5.5.1
• Applying mahalanobis() to obtain distances for observations after fitting the estimator. section 5.5.1
• Using cluster-relative distances when the data contains multiple clusters with different shapes or sizes. section 7.8.1

Robust covariance and distance calculation

from sklearn.covariance import MinCovDet cov = MinCovDet(random_state=0).fit(X) distances = cov.mahalanobis(X) Here, MinCovDet estimates the covariance structure and center, while mahalanobis() returns each observation’s covariance-adjusted distance. section 5.5.1

How Outlier Detection in Python covers it

How deep it goes

Explains geometry, inverse covariance, assumptions, robust estimation, and cluster-relative scoring, but does not derive the formula fully.

Examples

Demonstrates MinCovDet on three correlated features and applies mahalanobis() to observations.

What it assumes

Assumes familiarity with distance measures and covariance while introducing covariance-aware geometry and robust estimation.

What you can do after

Diagnose outliers with scikit-learn covariance estimators and extend scoring to observations relative to fitted clusters.

How it connects

Connects covariance matrices, Euclidean distance, principal component analysis, Gaussian distributions, clustering, and outlier detection.

liveBookmanning.com

Read Outlier Detection in Python on liveBook Get the book at manning.com

Related concepts

• distancekind of · Distance is a measure of separation between points or entities in a defined space or context.
• kernel density estimatealternative to · A kernel density estimate is a nonparametric probability-density estimate formed by combining localized kernel functions centered on observed data points.
• covariance matrixuses · A covariance matrix is a square matrix whose diagonal entries are variances and whose off-diagonal entries are pairwise covariances among variables.
• principal component analysisused by · Principal component analysis is a linear dimensionality-reduction technique that transforms data into orthogonal components ordered by the variance they capture.
• cluster-based local outlier factorcontrasts with · Cluster-based local outlier factor is an outlier detection algorithm that scores observations using cluster membership, cluster size, and distance to relevant clusters.
• Euclidean distancecontrasts with · Euclidean distance is the straight-line distance between two points in Euclidean space, calculated as the square root of the sum of squared differences between corresponding coordinates.

This explainer was generated by Manning's knowledge graph from the text of Outlier Detection in Python. The section links point to the parts of the book it draws on.