Practical Predictive Analytics

Description of the plots

Residual versus fitted: The residual versus fitted plot helps with determining if there are any patterns in the residuals. Ideally, the pattern (or ideally, no pattern) will be random scatter of the residual around a horizontal line.

If there is any obvious pattern, it is a good idea to revisit some of the bivariate correlations, or partial regression plot to try to tie the area back to some of the original scatterplots:

• Normal Q-Q: This plot will indicate if the residuals are normally distributed. It is a good sign if the residuals track the straight line.
• Scale location: This plot is used to check for the assumption of equal variances. It is also a good sign if this plot shows if residuals are spread equally along ...