What's the recommended approach for identifying new predictors when building gradient boosting or support vector machine predictive models? I know that visualizations are often used to come up with new predictors and/or transformations for linear models because linear models aren't able to capture non-linear behaviors or interactions in the same way that gradient boosting can. Are there visualization techniques that provide insight that you wouldn't get by just adding the candidate predictor to the model and seeing if it improves out-of-sample error?

I'm just using gradient boosting as an example, but I'm actually interested in visualization techniques that could be applied to other powerful models like support vector machines too.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.