Framework for imbalanced data classification
Mikołaj Błaszczyk, Joanna Jędrzejowicz
Abstract
Classifying imbalanced data remains a challenging task. The paper presents a framework for imbalanced datasets classification which makes use of different methods of oversampling and methods of dynamical selection of classifiers. The framework allows to perform extensive experiments to determine best possible configuration for the examined dataset in terms of geometric mean metric (g-mean). The results on benchmark datasets are presented.
Topics & Concepts
Computer scienceOversamplingBenchmark (surveying)Metric (unit)Task (project management)Machine learningSelection (genetic algorithm)Artificial intelligenceData miningPattern recognition (psychology)ManagementBandwidth (computing)GeodesyComputer networkOperations managementGeographyEconomicsImbalanced Data Classification TechniquesAnomaly Detection Techniques and ApplicationsStatistical and Computational Modeling