For more accuracy while fitting a linear model, we drop the input variables/features which are not significant(i.e pvalue>0.05) and whose variation inflation factor is  greater  than 5,I have landed up in a case where both input variables are highly correlated with each other and significant too,so which one should I drop?

Why is scaling for numeric variables done before splitting it into a train & test dataset?

Why do we need an intercept in a linear regression model?

If we use statsmodel.OLS do we need to add the intercept explicitly and how do we do it?

I was working on Dataset of Insurance in Python it had both categorical and numeric variables. For fitting a linear regression model I did the conversion of categorical to dummy variable, did scaling of whole data frame afterwards, after splitting training and testing data and, fitting model on training data, I found that VIF of one variable was more than 5 .(So while finding method, for reducing this high VIF, I came across Recursive Feature elimination method, in which it detects essential variables and nonessential variables). That particular variable with high VIF came under essential, so my query is should drop that variable by criteria of high VIF or keep it as it is? Is it only criteria for dropping variable?

How do I know if my model is Overfitting or Underfitting?