# Differences between normalization and standarization in multiple regression

Consider the following question regarding multiple regression

1) Can someone explain why we have to transform dependent variable using log-transformation (Normalization) when appear positive skewed y variable in regression model?

2) After log-transformation whether do I need to standardize that y variable using min max scale or StandardScaler methods?

3) If independent variables also have skewed data then do I need to Normalization for that variables or standardizing is enough?

I'll go through your question one by one.

1) Can someone explain why we have to transform dependent variable using log-transformation (Normalization) when appear positive skewed y variable in regression model?

Not necessarily log transformations, any kind of transformation (square, square-root, log, Z-scores, you name it) necessary to make the distribution of your data look more "Normal" (i.e. Gaussian). That is because all mainstream (frequentist) statistical models rely on the normality assumption of data (and residuals). When data are not Normal enough, the computation of parameters such as confidence intervals, standard errors, and p values will be unreliable.

2) After log-transformation whether do I need to standardize that y variable using min max scale or StandardScaler methods?

That is not mandatory either. Sometimes it is useful to scale your dependent variable in a range such that all its likely values are "easy to reach" by the parameters of your predictive model.

3) If independent variables also have skewed data then do I need to Normalization for that variables or standardizing is enough?

Normalization is a technique that can work. There is no general rule though. Sometimes you can use min-max scaling, some other times Z-scores work better. Other times again you'd achieve better performance with some custom scaling technique that invented ad hoc.

• Thanks Leevo for your valuables thoughts.. – randunu galhena Jan 2 at 16:42