Base idea

• Imagine you have a group of individual learners who are trying to solve a problem. Each learner has its own strengths and weaknesses, and they may not agree on the best solution. How can you combine their opinions and get the most out of them? This is the idea behind stacking.

• Stacking is a technique that uses a combiner to merge the predictions of several individual learners or base learners. The combiner is also known as meta algorithm or super learner, because it learns from the other learners and produces a superior prediction.

• This technique was introduced by Leo Breiman in 1996, but it was considered as a black art until 2007, when Van der Laan, Polley, and Hubbard developed its theoretical background. They showed that stacking can achieve optimal performance when base learners present high variability and uncorrelated predicted values.

under certain conditions, and they provided practical guidelines for its implementation.