Introduction

<aside> 💡

Scaling adjusts the range of data to ensure all features fit within a specific interval.

</aside>

This is useful for maintaining relative differences between values while ensuring data fits within a defined range.

It’s commonly performed using min-max scaling or max-abs scaling.

Why does this matter?

Like other methods of data transformation, scaling data is important for several important reasons:

• Features with different ranges can dominate machine learning algorithms, leading to biased results;
• Many algorithms perform better or converge faster with scaled data;
• Neural networks often require input features to be on similar scales for effective training
• Gradient descent optimisation can be more stable with scaled features;

Without proper scaling, some features might inadvertently have more influence simply due to their larger magnitude, rather than their actual importance to the problem.

Content map

To learn more about scaling, read each of the following sections:

Min-max scaling

Max-abs scaling

Range adjustment

Relative preservation of values

Suitability for bounded datasets

Next element

Standardisation

Notes

<aside> 💡

For example, if one feature ranges from 0-1 while another ranges from 0-1000, the second feature will have a disproportionate impact on the model's decisions unless proper scaling is applied.

</aside>