Search

Multiple classifier systems (MCS s) constitute one of the most competitive paradigms for obtaining more accurate predictions in the field of machine learning. Systems of this type should be designed efficiently in all of their stages, from data preprocessing to multioutput decision fusion. In this article, we present a framework for utilizing the power of instance selection methods and the search capabilities of swarm intelligence to train learning models and to aggregate their decisions. The process consists of three steps: First, the essence of the complete training data set is captured in a reduced set via the application of intelligent data sampling. Second, the reduced set is used to train a group of heterogeneous classifiers using bagging and distance-based feature sampling. Finally, swarm intelligence techniques are applied to identify a pattern among multiple decisions to enhance the fusion process by assigning class-specific weights for each classifier. The proposed methodology yielded competitive results in experiments that were conducted on 25 benchmark datasets. The Matthews correlation coefficient (MCC) is regarded as the objective to be maximized by various nature-inspired metaheuristics, which include the moth-flame optimization algorithm (MFO), the grey wolf optimizer (GWO) and the whale optimization algorithm (WOA). • Proposing a framework for discussing the intersection between instance selection, ensemble learning and swarm intelligence. • Heterogeneous ensemble system considering two additional steps to increase the diversity of the set of classifiers: Bagging, Distance-based feature selection. • Swarm intelligence for combining multiple decisions. • Matthews Correlation Coefficient (MCC) to evaluate ensemble under the search capabilities of swarm intelligence. • Getting out more accuracy from the reduced data. [ABSTRACT FROM AUTHOR]