Standardization Results
| Row | Feature | Original | Mean | Std. deviation | Difference | Z-score | |Z| | Position | Outlier | Reversed |
|---|
Original and Standardized Values
Z-Score Distribution
Reusable Machine Learning Code
Calculator Inputs
Formula Used
A z-score measures distance from the mean. The distance uses standard deviation units. Positive scores sit above the mean.
Negative scores sit below the mean. A score of zero equals the mean. Larger absolute scores indicate more unusual observations.
How to Use
- Select the required calculation mode.
- Enter values or upload a CSV file.
- Choose population or sample deviation.
- Select automatic or manual statistics.
- Set precision and an outlier threshold.
- Calculate, review charts, then export results.
Worked Example
| Value | Mean | Population deviation | Z-score |
|---|---|---|---|
| 10 | 20 | 8.1650 | -1.2247 |
| 20 | 20 | 8.1650 | 0.0000 |
| 30 | 20 | 8.1650 | 1.2247 |
Machine Learning Guidance
Standardization helps scale-sensitive algorithms compare features fairly. Common examples include KNN, SVM, PCA, clustering, regression, and neural networks. Tree models usually need less scaling.
Fit mean and deviation using training data only. Reuse those statistics for validation and test data. This prevents information leakage during evaluation.
Standardization differs from min-max normalization. Z-scores do not force values into a fixed interval. Extreme values can still remain extreme.
Frequently Asked Questions
What does a z-score represent?
It represents an observation’s distance from the mean. The distance is measured in standard deviations.
Should I use sample or population deviation?
Use population deviation for a complete population. Use sample deviation when data estimates a broader population.
Why can identical values not be standardized?
Identical values have zero standard deviation. Dividing by zero makes z-scores undefined.
Is a large z-score always an outlier?
No threshold is universally correct. Many workflows inspect absolute scores above two or three.
Does standardization create a normal distribution?
No. It changes location and scale only. The original distribution shape remains generally unchanged.
Can I standardize multiple columns?
Yes. Each numeric feature should normally receive its own mean and standard deviation.
How should test data be standardized?
Use the training set’s mean and deviation. Never refit statistics using the test set.
What happens to the standardized mean?
Automatically standardized training data has a mean near zero. Minor differences can result from rounding.
When is min-max normalization preferable?
It is useful when bounded values are required. It can be sensitive to extreme observations.