Ensemble methods use multiple learning algorithms to obtain better predictive performance than could be obtained from any of the constituent learning algorithms alone. By combining classifiers, we are aiming at a more accurate classification decision at the expense of increased complexity. The question is whether a combination of classifiers is justified. There is much work done on ensemble methodologies, much of it focusing on bagging and boosting methods. This work has a two-fold aim:
– Survey the latest literature for the state-of-the-art approaches and create a benchmark for these
– Extend local based methods and dynamic ensemble methods that have shown great potential.
For the latter purpose, ideas stemming from social sciences (such as electoral systems, delegative democracies etc.) can be used as a basis for an algorithm.
Indicative bibliography: