Mahalanobis Distance Calculator

Enter the point, mean, and covariance values to compute Mahalanobis distance.

How it is calculated

For a two-variable point x = (x1, x2), mean m = (m1, m2) and covariance matrix S = [[s11, s12], [s12, s22]], the tool computes D² = dᵀ S⁻¹ d with d = x − m, then D = √D². The 2×2 inverse is built from the determinant det = s11·s22 − s12²: S⁻¹ = (1/det) · [[s22, −s12], [−s12, s11]]. s11 and s22 are the variances of each variable, s12 is their covariance, all in the squared units of the data. Both D and D² are shown with four decimals. The covariance matrix must be positive definite: s11 and s22 above zero and a determinant above zero, that is s12² < s11·s22. Otherwise (including perfectly correlated variables) no result is shown. All seven fields are required.

Worked example

Point (5, 8), mean (3, 5), s11 = 4, s22 = 9, s12 = 3. The deviation is d = (2, 3). The determinant is 4·9 − 3² = 27, so S⁻¹ = (1/27)·[[9, −3], [−3, 4]]. Then dᵀS⁻¹d = (9·4 − 2·3·2·3 + 4·9)/27 = (36 − 36 + 36)/27 = 1.3333. This is D², and its square root gives a Mahalanobis distance of 1.1547.

Try this example →

Common mistakes

  • Entering standard deviations in s11 and s22. These fields take variances (standard deviation squared); a standard deviation of 2 means s11 = 4.
  • Entering the correlation coefficient (between −1 and 1) as s12. The covariance is the correlation multiplied by both standard deviations.
  • Using an impossible covariance matrix. It needs s12² < s11·s22; with s11 = 4 and s22 = 9, any s12 of 6 or more in size (determinant zero or negative) gives no result.
  • Dropping the sign of the covariance. With the same point and mean, s12 = 3 gives D = 1.1547 but s12 = −3 gives D = 2.0000, because the direction of the correlation changes which points count as typical.
  • Reading D as D². The tool shows both, and a chi-square cut-off for two variables applies to D², not D.

About this calculator

The Mahalanobis distance measures how far a point is from a distribution's mean, but unlike ordinary Euclidean distance it accounts for the correlation and spread between variables by dividing through the distribution's covariance matrix: D² = (x − μ)ᵀ Σ⁻¹ (x − μ). A point that's far along a direction where the data varies a lot is treated as less unusual than the same distance along a direction where the data barely varies.

This makes it a standard tool in multivariate outlier and anomaly detection, cluster analysis, quality control, and classification — for example flagging fraudulent transactions, detecting faulty sensor readings, or measuring how typical a new observation is relative to a training dataset in machine learning pipelines.

This calculator takes a two-variable point, the distribution's mean, and its 2×2 covariance matrix, then computes the Mahalanobis distance directly. Statisticians, data scientists, and students learning multivariate statistics use it to verify hand calculations or sanity-check outputs from statistical software.

Was this helpful?

You can pick up to 3 reactions.

Note from the editors

This version handles two variables, and it returns nothing when the covariance matrix is not positive definite, for example when the covariance is as large as the variances allow. For a rough outlier check on roughly normal data, compare the squared distance with a chi-square cutoff for two degrees of freedom; 5,991 marks the 95 percent level. Would a three-variable version be useful to you?

Comments

  • Be the first to comment.

Popular calculators

All Calculators