Hands-On Data Analysis with Pandas

Dummy classifier

This classifier will give us a model that is equivalent to the baseline we have been drawing on our ROC curves. The results will be poor on purpose. We will never use this classifier to actually make predictions; rather, we can use it to see whether the models we are building are better than random guessing strategies. In the dummy module, scikit-learn provides the DummyClassifier class precisely for this purpose.

Some interesting options for the strategy parameter of the DummyClassifier class are as follows:

• uniform: The classifier will guess each time whether or not the observation belongs to a hacking attempt.
• most_frequent: The classifier will always predict the most frequent label, which, in our case, will result in ...