I took 200 training points (100 per class) initially to run a random forest classification. When I run the last line of the code (below) to remove NA values, it removes NA values and 39 training points together from my data set.
### extracting raster values at training point positions
training_data