Nodes for regression

Hi,

I’ve been looking into using KNIME for regression analysis and noticed that the Linear Regression node provides only some basic measures in the output such as R-squared, std err, t-value and P>t.

Can anyone show me where I can find nodes with additional measures such as for the following:

  • Multicollinearity (VIF)
  • Model selection (AIC)
  • Independence of errors (Durbin Watson)
  • Homoscedasticity (Breusch Pagan)
  • Normality (Normal Q-Q)
  • Outliers (Cook’s Distance, Bonferroni Test)
  • Confidence intervals

Any help with this would be much appreciated.

 

I would probably look at our R integration and the "car" package.  It seems to have most of what you are looking for. 

http://cran.r-project.org/web/packages/car/car.pdf

Thanks Aaron for your response.  If I understand this correctly, I would use one of the R nodes in KNIME to connect to R where I would use the "car" package to do the analysis and then bring the output from R back into KNIME?

Yes, something like that.  We have a number of R nodes available that have various input and output ports.  Have a look at our youtube video on the subject.  If you post some simple example data I would be happy to try to prototype something for you.

https://www.youtube.com/watch?v=Dfm4-RuABm

Aaron,

I've attached a file with simple example data.  It's got an output variable plus three independent variables that are potential candidates for inclusion in the model.  If you're able to prototype something that uses R nodes to bring back into KNIME some of the measures mentioned earlier in this thread (such as VIF and AIC), it would be much appreciated.

By the way, I clicked on the youtube link but got a message that said "This video does not exist".

 

For those ones trying to calculate Variance Inflation Factor (VIF) in KNIME, you can use the following component:

It calculated VIF for all numeric variables in a table and filters out columns with VIF higher than a given threshold, to minimize collinearity.

3 Likes

Hi AdM.

It seems error when I drag VIF node from KNIME Hub to my workflow. The message said:


Unknown URL dropped!

URL can’t be dropped to KNIME workbench!

Unknown URL:

Hi @Long and welcome to the forum.

Are you by chance on a company network or VPN? I’m able to drag-and-drop from the Hub onto the workflow editor, so I’m wondering if that particular URL is being blocked on your side for some reason.

1 Like

@ScottF Thanks, you’re right. And I solved the problem.

2 Likes