I meant that this topic is full of ideas about the fusion of S1 and S2. There are multiple ways to do it, depending on your aim and the type of analysis which is performed. But if you want to perform a supervised classification, a stack containing all input sources is the best choice. Please have a look at these hints: Supervised classification for sentinel 1.
Another option is a principal component analysis or unsupervised classification but both require to scale the input data over the same value range first.