Like for example, I have my X_train, y_train and X_test. But y_test is left for prediction. We can perform prediction by applying classifier algorithm easily, however can we ever evaluate our model, now that we don't have any actual test data? Or if possible, how can we achieve the same?
How can we calculate metrics of a model when we don't have actual values for test data?
Test data should include the same variables as train data. Ideally, it should be a random selection from the whole data set.
So, what you can do is divide what you are currently calling train data into two sets and use one of those sets as test data.