Constructive meta-learning with machine learning method repositories

Here is discussed what is constructive meta-leaming and how it goes well compared with selective meta-learning that already becomes popular. Selective meta-leaming takes multiple learning schemes with the following different ways: bagging, boosting, cascading and stacking methods. On the other hand, constructive meta-leaming constructs the learning scheme proper to a given data set. We have implemented constructive meta-learning by recomposing methods into learning schemes with mining (inductive learning) method repositories that come from decomposition of popular mining algorithms. To evaluate our constructive meta-leaming, we have done the comparison of the performances of our constructive meta-learning and those of two stacking methods, using UCI/ML common data sets. It has shown us that our constructive meta-learning goes better than the two stacking methods. Furthermore, it turns out to be promising that we apply constructive meta-learning to meta-learner in selective meta-learning.