A dynamic multiple classifier system using graph neural network for high dimensional overlapped data

Mariana de Araujo Souza, Robert Sabourin, George D. C. Cavalcanti, Rafael M. O. Cruz

Abstract: Highlights•The local data distribution is crucial for most dynamic ensemble selection methods.•Class overlap and high dimensionality can negatively impact the locality definition.•We propose a technique that uses a multilabel graph neural network as a meta-learner.•The network learns from the local information and the classifiers’ interdependencies.•The method outperformed several techniques especially on challenging local scenarios.