Help: Random forest Parameters

Hello,
I am doing a land-use classification with RF in QGis (OTB). I have doubts about the choice of parameter numbers. Can someone help me?

I have 10 Sentinel 2 bands and 9 land use classes. How can I choose my number:
Maximum tree depth;
Minimum number of samples at each node;
Group the possible values of a categorical variable in k <= cat clusters to find a suboptimal division;
Size of the resource subset randomly selected on each node in the tree
Random seed.
Thank you so much for helping.
Gratitude.

Hi crismeries, when we talk about classification its basically different decision trees trying to learn with variety and different parameters in place. Let’s first try and answer below questions first and then explore more on the grid search and cross validation techniques to explore and identify the correct hyperparameters.

Is my distribution not the different classes same if not how is.random.forest classifier going to identify all ? Are we assigning any weights or keeping equal proportions of data by using upsampling or down sampling.

Shall we try some boosting classifier along and explore.more.on.those lines ?

Share your feedback on.the.above.and let me know if it helps :smile:

© Copyright 2013-2019 Analytics Vidhya