Residual Plot for regression

regression

#1

Hi,
For regression we look at residual plot on training set. If it is random (no pattern) then only we accept the fitness of model.What is the reasoning behind this. My understanding is that if its linear and residual has some pattern which means that one or more independent variables actually are not linearly correlated with dependent variable hence we go for feature engineering. can someone comment on this.

Regards,
Mithilesh