Random forest classifies the entire image into single class


I am using the random forest classifier in SNAP to classify a sentinel 2 image. But, it looks like the entire image gets classified into a single class.
The steps I followed:

  1. Downloaded S2B_MSIL2A_20210704T214049_N0301_R129_T09XVD_20210704T231052
  2. Resampled to 20m.
  3. Created training samples.
  4. Reprojected the image.
  5. Random forest classification.
    Although I selected all the bands as features, the classifier seemed to take only band 12 into account. It would be really helpful if someone could shed a light on what I am doing wrong. Thank you!

How many samples did you create per class? Are these equally distributed over the image and have similar proportions?
Also, did you actively select the bands B1 to B12 in the classifier. After resampling, there might be some quality layers inside which do not really contribute to classification.
Does the image look alright after reprojection?