Cost-effective ensemble models selection using deep reinforcement learning

Yoni Birman, Shaked Hindi, Gilad Katz, Asaf Shabtai

Abstract: Highlights•A cost-effective approach (SPIREL) for the selection of ensemble subsets.•SPIREL uses DRL to dynamically balance cost and performance.•SPIREL performs well in dynamic environments with partial information.•A highly generic approach which easily adapts to addition/removal of classifiers.•Can be easily trained on one dataset and applied to another (transfer learning).