Many machine learning practitioners talk about the need to transform the features & target variables, claiming this would boost the model accuracy.

Like this typical page elaborating about it in detail https://www.analyticsvidhya.com/blog/2020/07/types-of-feature-transformation-and-scaling/

I tried the techniques on several models using Wolfram V13. However, it didn't seem to result in any improvement to the model accuracy.

I noticed there is this RecalibrationFunction[] built-in to Classfy[] & Predict[] as a "post-processing" function, which would automatically correct overconfident or underconfident classifiers. The documentation doesn't explain much about how it is working in the background. Is this RecalibrationFunction[] in fact doing something similar to the data pre-processing, such as log transformation, scalar transformation?

Thanks