NeuroNomixer logoNeuroNomixer
  • Home
  • Blog
  • Visual Guides
  • Authors
  • Contact
Sign InSign Up
NeuroNomixer logo
HomeBlogAuthorsContactPrivacy Policy

© 2026 NeuroNomixer — Built with Next.js & Tailwind CSS

Visual Guides/Outlier Detection
Data & Analysis

Outlier Detection: Spot the Odd One Out

Drag, add, or remove data points and watch the mean, median, and trend lines update when you drop them. Toggle between Z-Score and IQR methods to see which points get flagged, and why the answers can differ.

Z-Score explored
IQR explored
Point dragged

Sign in to track progress

Z-Score Method: flagging points more than 3σ from the mean on either axis · 0 outliers detected

15 points · 0 flagged

02020404060608080100100X ValueY Value(48, 52)(51, 48)(50, 51)(49, 49)(52, 50)(50, 47)(51, 52)(48, 51)(50, 50)(49, 52)(52, 48)(75, 30)(20, 80)(85, 15)(95, 5)NormalOutlierMean XMedian XTrend

Drag to move · click empty area to add · double-click to remove

Z-Score Method

Measures how many standard deviations each coordinate is from that axis mean, applied to X and Y separately. Assumes a roughly normal distribution. A key limitation: extreme outliers inflate the standard deviation, which can actually mask themselves.

Best for

Normally distributed data

Formula

z = (x − μ) / σ

Limitation

Outliers inflate σ, masking themselves

Used in

Quality control, anomaly detection

Detection Method

Threshold (σ)3.0σ

Mean (X / Y)

55.0 / 45.3

Std Dev (X / Y)

17.2 / 16.8

X fences

3.5 to 106.5

Y fences

-5.2 to 95.9

Flags points more than 3.0σ from the mean on either axis. Assumes a roughly normal distribution per axis.

Flagged Outliers

0 / 15

No outliers detected with current settings

Live Statistics (X-axis)

Mean X

55.0

sensitive to outliers

Median X

50.0

robust to outliers

Std Dev X

17.2

spread of values

N points

15

0 flagged

No outliers flagged: mean and median are both reliable.

When to Use Which

Z-Score: Data is normally distributed; you need a threshold in interpretable σ units.
IQR: Data is skewed or you can't assume normality; need resistance to masking.
Both agree: High confidence the point is genuinely anomalous.
They disagree: Use domain knowledge: the disagreement reveals the distribution shape.
← Previous GuideNext: Correlation vs Causation →