Strictly, at least one band as the maximum likelihood classification can be formulated for scalars (one band) or vectors (multiple bands). The maximum likelihood estimator just finds the value of x that maximises the probability density function P(y|x), where P(y|x) dy is the probability that y lies in (y, y+dy) when x has a given value. X and y can both be scalars or vectors. Whether the classification will be good or not depends on the  relation between the measurements and the variables to be estimated from them. There is no general answer to the question unless the problem is more specifically posed. 

I used three channels.

http://lipovscak.com/radovi/1986_iete.pdf