1

I am interested in having an approximate idea of the different sensitivity of regression methods to outliers. A kind of "user guide" or manual to use.

I know linear regression is sensitive to outliers, and I suppose this is also valid to non-linear regression (am I right?). I also know that boosting methods are sensitive, too. And I have read that neural networks are relatively robust to outliers. What about Support Vector Regression?

Could you please confirm these and give an intuition of the sensitive of different regression methods to outliers? It would be nice to have a kind of ordered list of methods from more sensitive to less sensitive: method_1 > method_2 > method_3... if this is possible (that perhaps not).

Thanks.

David
  • 271
  • 1
  • 6
  • 5
    All regression methods based on minimization of a convex loss function will be sensitive by outliers in the design space. There are nice review papers for specific types of regression models (for example [this](https://arxiv.org/pdf/1510.01064.pdf) one for classification methods). – user603 Nov 16 '17 at 12:47
  • 2
    Possible duplicate of [Interpreting case influence statistics (leverage, studentized residuals, and Cook's distance)](https://stats.stackexchange.com/questions/138670/interpreting-case-influence-statistics-leverage-studentized-residuals-and-coo) – kjetil b halvorsen Jul 25 '18 at 17:49

0 Answers0