Your friend's Mini$-$Project group fit a classifier model on a dataset taken from a Kaggle repository. using an $80\frac{\%}{20}**$ train$-$test split. In their final project report, they declared this model as the best performer due to ite yery high training and test accuracies, Later, however, you discovered another dataset in the Kagyle repository that was meant fo be used as the test data for their model. When you tried their model on this new test dataset, you were surprised to observe very low prediction accuracy.

List at least three different possible causes for the low accuracy observed on the new dataset. How could the group have protected itself from these poor outcomes? Be concise.