What is 'greedy' feature selection exactly?

kokszoskaroj

Hi, 

in linear regression there is a feature selection option 'greedy'. How could be this selection method described?
I would also like to know about the a mathematical background of it. Are there any papers that describe this method precisely? In addition I'd also liked to know whether it can be used for time series regression (i'm using windowed data). 

Thanks, 
Daniel

varunm1

Hi @kokszoskaroj

A greedy feature selection is the one in which an algorithm will either select the best features one by one (forward selection) or removes worst feature one by one (backward selection). 

There are multiple greedy algorithms. In rapidminer, the greedy algorithm used is described in the below link.

https://docs.rapidminer.com/latest/studio/operators/modeling/optimization/feature_selection/optimize_selection.html

Hope this helps.

varunm1

Hi @kokszoskaroj

A greedy feature selection is the one in which an algorithm will either select the best features one by one (forward selection) or removes worst feature one by one (backward selection). 

There are multiple greedy algorithms. In rapidminer, the greedy algorithm used is described in the below link.

https://docs.rapidminer.com/latest/studio/operators/modeling/optimization/feature_selection/optimize_selection.html

Hope this helps.

amitd

Are there any additional details on how the 'greedy' option in the Linear Regression operator?