When should I do feature scaling or normalisation in machine learning?

Question

I have a training feature set consisting of 92 features. Out of which 91 features are boolean values of 1 or 0. But 1 feature is numerical and it varies from 3-2000.

Will it be better if I do feature scaling on my 92nd feature?

If yes, what are the best possible ways to do it? I am using Python.

if using python then use minmax scaller from sklearn.preprocessing, you will be able to scale the data from 0,1 in the particular array that has huge values. — Inder, Aug 11 '18 at 17:30
http://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.MinMaxScaler.html?qt1m4dc=1 — Inder, Aug 11 '18 at 17:30
Which algorithm are you planning to use? Also does feature 92 have outliers? — Dimosthenis, Aug 11 '18 at 20:40

score 0 · Accepted Answer · answered Aug 12 '18 at 12:49

Sometimes, It is highly dependent on which algorithm you wanna use for your prediction. Suppose if you are using SVM and using Gaussian Kernel for that and you are not using feature scaling on your inputs then you might end up with wrong hypothesis and your large features will dominates over the other smaller features. Generally, feature scaling are always the best ways to control the variations in input and also it makes algorithm to compute fast (or in other words converge to the optimal minima).

When should I do feature scaling or normalisation in machine learning?

1 Answers1