That's really unfortunate; results are straight out of the algorithm but it's reflective of a bias in the training data set.