RF classification

Hi, i am doing a RF supervised clasification with the periphery of tunisia. What can I do to make the classes more specific?
In many areas, like the ones in the image, the vectors I made are confused

looking at the area in general the classes seems correctly placed, but when I zoom I get confused pixels

You can include additional images from different times of the year.

I have tried it, and this August photo is the one that is most correct.
Maybe is a problem with the size of the vectors?

It is normal that there is confusion between classes. If you want to improve your results you need to add images where there are spectral differences between the classes in question.

You should try using several input images at the same time, from different times of the year/growing season.