Open Problems

• 1) when to use which ensemble methods - Adaboost best except when there is noisy data
• 2) Bagging and ECOC combined perform better than either separately - other combinations should be explored
• 3) few systematic studies of ensembles on ANN and rule-learning systems
• 4) ensembles take a lot of memory - can they be converted to less redundant representations
• 5) ensembles provide little insight - can we obtain explanations from ensembles