Makes sense, statistically smaller sample to be trained on, relatively easy fix, just retrain with more motorcycles in the data.