---
title: "Comment by Mark on \"Demystifying Feature Scaling\" | Hashnode"
description: "This is an interesting topic. It&#39;s essentially a way to compare &#39;incomparible&#39; dimensions. I.e. you might have the top speed of a car, the weight an"
url: "https://hashnode.com/posts/demystifying-feature-scaling/5cfa3d6d7558204d13bf6886/comment/5cfa5bc9145845623003178a"
---

[Discussion](https://hashnode.com/posts/demystifying-feature-scaling/5cfa3d6d7558204d13bf6886)

Comment [Demystifying Feature Scaling](https://hashnode.com/posts/demystifying-feature-scaling/5cfa3d6d7558204d13bf6886)

[Mark](https://hashnode.com/@mverleg)

[Jun 7, 2019](https://hashnode.com/posts/demystifying-feature-scaling/5cfa3d6d7558204d13bf6886/comment/5cfa5bc9145845623003178a)

This is an interesting topic. It's essentially a way to compare 'incomparible' dimensions.

I.e. you might have the top speed of a car, the weight and the horse power of the engine, and you want to determine if it's similar to another car. But if car A and B differ in weight by 100kg and car A has 2 more horse power than C, then is A more similar to B or to C?

Standardizing is an approach that let's you not think about the conversion factor from kgs-of-car to hps-of-engine. It's not always the best approach, but it's usually the easiest, and maybe less likely to have human bias.

> any Algorithm which is Not Distance based is Not affected by Feature Scaling

Also many non-linear things like neural networks (assuming a non-linear activation function).
