Thakshashila: Created page with "= Normalization (Machine Learning) = '''Normalization''' in machine learning is a data preprocessing technique used to scale input features so they fall within a similar range, typically between 0 and 1. This helps improve model performance, especially for algorithms sensitive to the scale of data. == Why Normalize Data? == Some machine learning algorithms (e.g., K-Nearest Neighbors, Gradient Descent-based models, Neural Networks) perform better when input features ar..."

2025-06-10T06:34:09Z

Created page with "= Normalization (Machine Learning) = '''Normalization''' in machine learning is a data preprocessing technique used to scale input features so they fall within a similar range, typically between 0 and 1. This helps improve model performance, especially for algorithms sensitive to the scale of data. == Why Normalize Data? == Some machine learning algorithms (e.g., K-Nearest Neighbors, Gradient Descent-based models, Neural Networks) perform better when input features ar..."

New page

= Normalization (Machine Learning) =

'''Normalization''' in machine learning is a data preprocessing technique used to scale input features so they fall within a similar range, typically between 0 and 1. This helps improve model performance, especially for algorithms sensitive to the scale of data.

== Why Normalize Data? ==

Some machine learning algorithms (e.g., K-Nearest Neighbors, Gradient Descent-based models, Neural Networks) perform better when input features are on a similar scale. Without normalization, features with larger numeric ranges may dominate others, leading to biased results.

== Common Normalization Techniques ==

=== 1. Min-Max Normalization ===

Scales features to a fixed range, usually [0, 1].

:<math>
x' = \frac{x - x_{\text{min}}}{x_{\text{max}} - x_{\text{min}}}
</math>

* Best for bounded data.
* Sensitive to outliers.

=== 2. Z-score Normalization (Standardization) ===

Centers the data around the mean with unit variance.

:<math>
x' = \frac{x - \mu}{\sigma}
</math>

Where:
* <math>\mu</math> = mean of the feature
* <math>\sigma</math> = standard deviation

* Useful for algorithms that assume Gaussian distribution.

=== 3. Max Abs Scaling ===

Scales data by dividing by the maximum absolute value:

:<math>
x' = \frac{x}{|x_{\text{max}}|}
</math>

* Preserves zero entries in sparse data.

=== 4. Robust Scaling ===

Uses median and interquartile range (IQR) to scale:

:<math>
x' = \frac{x - \text{median}}{\text{IQR}}
</math>

* Less sensitive to outliers.

== When to Use Normalization ==

Use normalization when:
* Input features are measured in different units or ranges.
* You use distance-based algorithms (e.g., k-NN, SVM).
* You're training neural networks using gradient descent.

== When Not to Normalize ==

* When using tree-based algorithms like Decision Trees or Random Forests (these are insensitive to feature scale).
* When your features are already on the same scale or naturally bounded.

== Example ==

If feature A ranges from 1 to 1000 and feature B from 0 to 1:
* A normalized model ensures both features contribute equally to model training.
* Without normalization, feature A may dominate due to its larger range.

== Related Pages ==

* [[Feature Scaling]]
* [[Standardization]]
* [[Preprocessing (Machine Learning)]]
* [[K-Nearest Neighbors]]
* [[Gradient Descent]]

== SEO Keywords ==

normalization in machine learning, min max normalization, z score normalization, feature scaling ML, standardization vs normalization, when to normalize data, data preprocessing techniques

Normalization (Machine Learning) - Revision history