Basic regressionquestion

#1
Hi,
I am currently learning ML algorithms and implementing in R. I have a some basic questions.
1.)I know that linear regression an statistical model where prior to building model it should satisfy some of assumptions(Hypothesis) like
>>All the attributes in the dataset must be IID.
>>Residuals must be normally distributed.
>>Homoscedasticity among attributes.
How do i check if attributes satisfy these assumptions prior to building model.
>>Does doing cor() on attributes and removing the attributes with higher correlation assure my attributes are Homescedastic.
>>Regarding I.I.D do i need to do t.test() or chi.square among all the attributes?
I may be wrong in many ways please correct me.
Sorry, if this is an naive question.
Thanks
 

noetsi

Fortran must die
#3
You check if the residuals are normally distributed best by putting the residuals on a QQ plot. Although there are formal test of homoscedasticity probably the best way is to simply look at the residuals against the predicted values and see if a pattern exist. There should not be one if the assumption of Heteroscedasticity is met. Other assumptions that should be tested are non-linearity, partial regression plots are best for that. Again there probably should not be a pattern. There is no test I am aware of for independence. If you design your analysis correctly this should not occur. An exception is autocorrelation with time series, there are a variety of test for it including Durbin Watson, although best known it is not the best test.