Is the random forest solution for regression interpretable and sparse? I have a regression problem scenario. Basically, I want to model a certain biological problem with regression models and at the end my model should be interpretable.
I need to have a sparse model. So I'm trying Lasso and Elastic Nets. But the performance of these methods in my data set is not good. 
I'm thinking about using a random forest for the regression. 


*

*Would someone give me the name of the most useful R package for random forest for regression? 

*Also, is the random forest solution for regression sparse and interpretable? 

 A: Random Forest fits a very large number of individual decision trees. These trees then "vote" to obtain a final output for the entire forest. Directly interpreting a decision trees can be easy or tricky (depending on the # of variables and correlations), but a large set of them at once is virtually impossible. So no, Random Forest is much less interpretable than methods like Lasso and Elastic Net. It's also not sparse, because it tends to utilize all variables. 
To make the reasons behind this more intuitive:
personally I like to imagine Random Forest as trying to see the problem from all possible different perspectives offered by your data. The perspectives can be very different from each other, but when they agree on an answer you can be reasonably confident that the answer is correct. This perception of Random Forest also helps answer your question. You can see and understand the problem from one perspective but not from all possible ones at the same time, so you cannot easily understand what your Random Forest does. Also all possible perspectives include all the features, so it also cannot be sparse.  
