Calculator Inputs
Example Data Table
This example uses sample deviation, a two-tailed rule, and a threshold of 2.5.
| # | Value | Approx. Z Score | Status |
|---|---|---|---|
| 1 | 12 | -0.57 | Inlier |
| 2 | 13 | -0.46 | Inlier |
| 3 | 13 | -0.46 | Inlier |
| 4 | 14 | -0.35 | Inlier |
| 5 | 14 | -0.35 | Inlier |
| 6 | 15 | -0.24 | Inlier |
| 7 | 15 | -0.24 | Inlier |
| 8 | 16 | -0.14 | Inlier |
| 9 | 16 | -0.14 | Inlier |
| 10 | 17 | -0.03 | Inlier |
| 11 | 45 | 2.98 | Outlier |
Formula Used
The calculator standardizes each observation and compares it with your chosen threshold.
1) Mean
mean = (sum of all values) / n
2) Population Standard Deviation
sigma = sqrt( sum((x - mean)^2) / n )
3) Sample Standard Deviation
s = sqrt( sum((x - mean)^2) / (n - 1) )
4) Z Score
z = (x - reference_mean) / reference_standard_deviation
5) Outlier Decision
Two-tailed: abs(z) >= threshold
Upper-tail: z >= threshold
Lower-tail: z <= -threshold
In machine learning preprocessing, this helps identify unusually large or small feature values that may distort scaling, training stability, and model interpretation.
How to Use This Calculator
- Enter a dataset label for easier reporting.
- Paste numeric values separated by commas, spaces, or new lines.
- Choose a z score threshold such as 2.5 or 3.0.
- Select sample or population deviation.
- Choose two-tailed, upper-tail, or lower-tail screening.
- Use dataset statistics or supply a custom mean and deviation.
- Click Calculate Outliers to show the summary and detailed table.
- Use the CSV or PDF buttons to export the results.
Why This Matters in AI & Machine Learning
Outlier screening can improve feature consistency before normalization, clustering, regression, classification, or anomaly detection workflows. A z score method is most useful when data is reasonably centered and variance is meaningful. It is fast, interpretable, and easy to audit.
For skewed, heavy-tailed, or non-Gaussian variables, combine this review with box plots, IQR screening, robust scaling, winsorization, or model-specific anomaly checks.
FAQs
1) What does a z score outlier calculator do?
It measures how far each value sits from a reference mean in standard deviation units. Values beyond your threshold are marked as potential outliers for review.
2) Which threshold should I use?
Common thresholds are 2.5 or 3.0. Lower thresholds flag more values, while higher thresholds are stricter and usually isolate only the most extreme observations.
3) When should I use sample deviation?
Use sample deviation when your data is a subset drawn from a larger population. It is often preferred for exploratory modeling datasets and training samples.
4) When should I use population deviation?
Use population deviation when the dataset represents the full group you want to analyze. It treats your entered values as the entire population.
5) What is the difference between two-tailed and one-tailed screening?
Two-tailed screening checks unusually high and low values. Upper-tail checks only very large values, while lower-tail checks only very small values.
6) Can z score outlier detection fail?
Yes. It can be misleading for highly skewed data, tiny samples, or variables with strong non-normal behavior. Use domain knowledge and complementary checks.
7) Why would I use a custom mean and deviation?
Custom reference statistics help when you want to compare a new batch against training-set baselines, production benchmarks, or known process limits.
8) Should I always remove flagged outliers?
No. Some extreme points are valid signals. Review data quality, domain meaning, model sensitivity, and business impact before deleting or capping observations.