Z-Score Outlier Detection & Data Cleaning

Dataset

Sample size (n) 400

Outlier fraction 8%

Z-Score Threshold

Threshold |z| > 2.5

Live Statistics

Flagged as outliers

0 / 0

Mean |z| of flagged

0.00

Std dev (raw)

0.00

Std dev (cleaned)

0.00

Corr(x,y) raw

0.00

Corr(x,y) cleaned

0.00

Legend

Inlier (|z| ≤ threshold on all axes)

Outlier (|z| > threshold on any axis)

How it works

Each point is a 3-feature sample (x, y, z). For every feature the simulator computes the population mean μ and standard deviation σ, then a per-axis Z-score for each sample:

z_i = (x_i - μ) / σ

Flag as outlier if  max(|z_x|, |z_y|, |z_z|) > threshold

This is the classic univariate outlier-detection rule used as a first-pass data-cleaning step before statistical modeling: any feature that lands more than a chosen number of standard deviations from its own mean gets flagged, independent of the other features.

Sample size / outlier fraction — regenerates a synthetic dataset: a correlated Gaussian "clean" cluster plus a fraction of points drawn from a wider, shifted distribution to simulate sensor glitches or bad records.
Threshold slider — moves the acceptance boundary; a lower threshold flags more points (more sensitive, more false positives), a higher one flags fewer (more conservative, may miss real anomalies). 2–3σ is the common rule-of-thumb range.
Hide flagged / Show all — toggles whether flagged points stay visible (dimmed red) or are removed from view entirely, mirroring the "drop outliers" step of a cleaning pipeline.
Correlation readouts — Pearson correlation between the x and y features, computed on the raw data and again after removing flagged points, showing how a handful of outliers can distort a statistical relationship even when most of the data is well-behaved.

Real-world relevance: this exact Z-score screen is a standard first step in data-science pipelines — before fitting a regression, clustering model or dashboard, analysts routinely flag and review points that fall far outside each feature's normal range.

Inlier Outlier

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.