Dear Nikhil,

Technically RF can support 27 classes, but the more class you use, the more training data you'll need to get the same accuracy/goodness-of-fit.

HTH,

Ákos

http://www.academia.edu/download/29631954/Rodriguez_Galiano__etal_2012_Random_forest_land_cover_ISPRS.pdf

Article Random forest classifier for remote sensing classification

I agree with Alessandro and Akos. Regards.

Dear All,

Thank you for your responses. They are very helpful.

Ákos Bede-Fazekas I trust that we have capability to gather more training data. Is there any rule of thumb on how much training data is needed for each class?