Eliminating Homoscedasticity from the data set before linear regression



Hello everyone,

There are several assumptions that we take before applying linear regression.
These include :

  1. Eliminating multicollinearity.
  2. Normalization of the variables
  3. Eliminating homoscedasticity that is the correlation between error terms and independent variables.

For the third point, how do we treat a data set that has homoscedasticity in it? And how significant a change do we observe by doing so in linear regression?


Hi @Prateek123

i means your relation is no linear therefore you could do transformation of variables (try to) , but I would look for a non linear methods, welcome to Tree !!!
Hope this hep.


You can go for weighted least square regression.


Thanks for your response. I still had a few doubts regarding it.
@Lesaffrea, does homoscedasticity imply non-linearity of the data set?

@anurag.monk The weighted least square regression algorithm assumes that the weights are known in advance. However, in reality the weights are unknown to us. Would this algorithm work well without prior knowledge of weights?