The most appropriate response is:
The dataset is skewed in one direction.
Biased results from algorithms often stem from the data they are trained on. If the dataset is not representative or is biased in some way, the algorithm will likely reproduce that bias in its outputs.