Tagging NA value in a continues variable I'm using Random Forest model to predict academic success. As far as I know RF doesn't work with NAs.
One variable is a grade in a certain course, which is not mandatory for the students, thus some don't have a grade at it.
Now, this variable is important to me because I want to find whether this specific course have any effect on the academic success. But, 0/1 dummy variable is not good enough for I need to distinguish between the grades of those who did participate.
So far, the only solution I can think of is to put "-1" value instead of the "NA", however, I never saw anyone doing such a thing - can it cause issues? How can I solve this problem? 
 A: I can see one possible issue with this approach since the Decision Trees will weigh the -1 value in relation to the other grades as if -1 actually was a grade. To clarify: Let's assume you have grades 


*

*2 = pass

*1 = fail

*-1 = not taken course


The difference (distance) between pass and fail is 2 - 1 = 1, but the difference between fail and no grade is 1 - (-1) = 2. Not taking the course can then be interpreted as "worse" than failing it, and also has a bigger impact than the difference whether people passed or failed. This might not be what you are looking for and thus problematic.
One way of solving this issue is to convert this to categorical variables e.g. pass/fail/not participating and one-hot-encode these. This, however, would introduce another problem in that the Decision Tree now can't make assumptions like pass > fail (2 > 1 above) since comparison between categorical values makes no sense. So, this might not be what you want either and thus problematic.
In this light, you have to choose the NAs values (or even value type) according to your problem (assumptions, hypothesis, other features etc) and be careful since the choice will have an effect on the end-results (and may  even introduce unintentional assumptions).
