Ok so I'm confused about the whole concept of standardizing data. I get the concept of why we need to standardize data for, let's say multiple linear regression so the data points are similar, but what I don't understand is how the model learns to predict from this. I am confused because my thought is that if we feed the ML model modified versions of the inputs, how is it going to accurately predict the actual inputs. For example, say we are predicting weight based on height and age. Age is in years and height is in inches so we need to standardize the data. Because of this the ML model trains on input data like: {height: 0.7, age: 1.2}. So how is it supposed to accurately predict the weight if it is given input data like this {height: 67, age: 14}, if it trained with scaled data. Any thoughts?
2
-
1$\begingroup$ ML? Machine Learning? Maximum Likelihood? Mixed Linear? The first time you use abbreviations in your posts, can you please specify what it is that you are abbreviating? $\endgroup$– StatsStudentCommented Apr 22, 2021 at 3:59
-
3$\begingroup$ Suppose the standardization converts weight from pounds to kilograms. You ask, "how is the model going to accurately predict when the weight is expressed in kilograms?" The answer is obvious: it predicts in kilograms. If you want to express the results in pounds, just convert the units back again. $\endgroup$– whuber ♦Commented Apr 22, 2021 at 13:14
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Well, if you train your model with scaled data, then you will have to do predictions with scaled data too. It is not a problem because you know the function you used to transform the original data and you can apply exactly the same for future (or test) data.
