I've found a significant number of papers that are employing Genetic Programming approaches such as symbolic regression (SR) to better model complex environmental systems. Many workers note that SR provides meaningful insight over the relative sensitivity and magnitude of response of multiple features used to estimate a given measure. However, it is my understanding that the sensitivity and magnitude of the response of multiple features within a traditional linear regression model is often unintuitive because of their dependence on all other features. Is the latter also an issue for SR type approaches?
In a traditional regression model, you (generally) do not look at interactions. From a traditional point of view, there are a lot of reasons for this. Those same issues can wreck havoc on most modeling methods.
With a linear regression, there is also an assumption that there is a linear dependence between X and Y.
With more modern approaches, the assumption about linearity is not required. You can find all sorts of useful interactions and non-linearities.
If you understand how traditional regression models work, you can improve them quite significantly.
Thanks Andrew,
Many workers note that SR provides meaningful insight over the relative sensitivity and magnitude of response of multiple features used to estimate a given measure. However, it is my understanding that the sensitivity and magnitude of the response of multiple features within a traditional linear regression model is often unintuitive because of their dependence on all other features. Is the latter also an issue for SR type approaches?
Kind regards,
Clint
- Badges
- Science topic
- Similar topics
- Papers