Normalize Data Transform

The Normalize Data transform rescales numerical columns for analysis or modeling.

Basic Usage

To rescale numerical data:

1. Select the Normalize Data transform from the transform menu.
2. Select one or more fields under Target Columns.
3. Choose a Method.
4. Configure any settings shown for the selected method.
5. Apply the transformation.

note

Target Columns lists numerical columns only.

Configuration Options

Basic Options

• Target Columns: Select one or more numerical columns to rescale.
• Method: Choose one of these scaling methods:
  • Min-Max Scaling
  • Z-Score Standardization
  • Robust Scaling
  • Max Absolute Scaling
  • Normalizer
  • Quantile Transformation
  • Power Transformation

tip

Hover over a method to see its short description.

Advanced Options

The workflow UI shows settings only for methods that need them.

Min-Max Scaling

• Range From: Lower bound (default: 0)
• Range To: Upper bound (default: 1)

Z-Score Standardization

There are no advanced Z-score settings in the UI. Standard execution always subtracts the mean and divides by the sample standard deviation (ddof=1).

Max Absolute Scaling

There are no advanced settings for this method. Each selected column is divided by its maximum absolute value.

Robust Scaling

• Quantile Range: IQR range used for scaling (default: 25-75)
• With Centering: Center the data first (default: True)
• With Scaling: Scale by the IQR (default: True)

Normalizer

• Norm: Norm used for normalization (options: 'l1', 'l2', 'max')

Quantile Transformation

• Number of Quantiles: Number of quantiles to compute. The workflow UI defaults to 1,000. Standard execution caps this value at the row count, so an input with fewer than 1,000 rows uses its row count. Direct backend calls that omit the parameter also default to the row count.
• Output Distribution: Output distribution (options: 'uniform', 'normal')

Power Transformation

• Method: Power method (options: 'yeo-johnson', 'box-cox')

Normalization Methods

Min-Max Scaling

Maps values into a fixed range, usually 0 to 1.

Use for: Bounded values or algorithms that require non-negative input.

Z-Score Standardization

Centers values at 0 and scales them to a standard deviation of 1.

Use for: Roughly Gaussian features measured on different scales.

Robust Scaling

Uses the median and quantile range, reducing sensitivity to outliers.

Use for: Data with outliers that would distort mean- or range-based scaling.

Max Absolute Scaling

Scales each feature by its maximum absolute value.

Use for: Sparse data or any case where zero must remain zero.

Normalizer

Scales each row to unit norm.

Use for: Comparing feature proportions rather than absolute magnitude, often in text classification or clustering.

Quantile Transformation

Maps values to a uniform or normal distribution.

Use for: Spreading frequent values or reducing outlier impact.

Power Transformation

Applies a power function to make a distribution more Gaussian-like.

Use for: Skewed data whose variance needs stabilizing.

Examples

Example: Normalizing Student Grades

Input Dataset:

Student_ID	Student_Name	Subject	Grade	Ranking
101	Alice	Math	0	6
102	Bob	Science	20	5
103	Charlie	English	40	4
104	David	Math	60	3
105	Eva	Science	80	2
106	Frank	Math	100	1

Configuration:

• Target Columns: Grade
• Method: Min-Max Scaling
• Range From: 0
• Range To: 1

Result:

Student_ID	Student_Name	Subject	Grade	Ranking
101	Alice	Math	0.0	6
102	Bob	Science	0.2	5
103	Charlie	English	0.4	4
104	David	Math	0.6	3
105	Eva	Science	0.8	2
106	Frank	Math	1.0	1

Best Practices

1. Match the method to the data distribution and model requirements.
2. Use Robust Scaling or another less-sensitive method when outliers are present.
3. Use Max Absolute Scaling when zero must stay zero.
4. Apply the same fitted scaling approach to training and test data.
5. Check the assumptions behind Z-Score and distribution-based methods.

Troubleshooting

• A field is missing: If it does not appear under Target Columns, check that it is numerical.
• Min-Max Scaling looks compressed: Check for extreme values.
• Box-Cox fails: It requires strictly positive values; zero and negative values are not accepted.