How can non linear relationships be converted into linear ones using transformations?

transformations
feature_engineering

#1

Hello,

As, we need to have a linear relationship between the variables to comprehend their relationship better, we need to transform non-linear relationships to linear ones. But I don’t seem to understand how this can be done. Can someone explain this with an example?
Suppose we see a curvilinear relationship between two variables using scatter plot, how can we get a linear relationship out of it? Does log transformation works in most of the situations? Do we need to try out all the transformations?
What would be a good way to tackle this issue while solving problems?

Thanks


#2

Hello @Ravi,

Log transformations is one of the ways in which this issue is dealt with.Like in the case of logistic regression the log model is used exactly because of this reason.However other transformations like sqrt,exponentiation also work at times.
However no data will be exactly linear but rather show a linear trend.
So you have to try out with different methods and see which one gives a better linearity.Also log transformations are a little different to interpret and hence that has to be taken care of if you are using log transform.
Hope this helps!!