What’s The Difference Between Scaling and Transformation?
Understanding the Key Differences Between Scaling and Transformation
When we prepare features to train machine learning models, there are 2 common procedures that we normally employ: variable scaling and variable transformation.
They are not the same, and more importantly ⇒ they serve different purposes!
👉 Scaling, in its essence, is about resizing while maintaining the original shape.
👉 It involves multiplying the values with a constant and shifting the distribution left or right.
No, scaling does NOT change the variable distribution.
👉 Transformation is about changing the variables’ distribution.
👉 We can do so by applying mathematical operations like logarithm and power transformations.
We change distributions to help meet certain model assumptions, in general, linear model assumptions. If the models make no assumptions on the data, then there’s no need to transform the variables.
🎁 Want to know more?
Variable transformation - Read this article
Variable scaling - Read this article
Our course on Feature Engineering
Our book on Python Feature Engineering Cookbook
Ready to enhance your skills?
Our specializations, courses and books are here to assist you:
Advanced Machine Learning (specialization)
Forecasting with Machine Learning (course)