Actually here kaggle wants the predictions on the given prediction data(depends on this predictions only we will get private leaderboard scores). So here I am taking the advantage of the test data also to train my models and to perform better on the prediction data.

But typically in the real world we will use test data for hyper-parameter tuning and prediction data for model evaluation.

I hope now it is clear :)