Resumen

VERDECIA-CABRERA, Alberto; FRIAS-BLANCO, Isvani; QUINTERO-DOMINGUEZ, Luis  y  SARABIA, Yanet Rodríguez. Learning with meta-classifiers from non-stationary data streams. Rev cuba cienc informat [online]. 2020, vol.14, n.4, pp. 20-33.  Epub 01-Dic-2020. ISSN 2227-1899.

Many sources generate large amounts of data constantly over time, which are known as data streams. Because of these are acquired over time and the dynamics of many real situations, the distribution of probabilities (target concept) that governs the data can change over time, a problem commonly called concept drift. This article presents a new algorithm based on classifiers ensembles for learning from data streams with possible concept drifts. The proposed algorithm uses meta-classifiers to combine the predictions of the base classifiers of the ensemble, and maintains a set of adaptive classifiers to manipulate possible concept drifts. The proposed method meets the common requirements for online learning from data streams: it is capable of processing input data with constant temporal and spatial complexity, and also only processes each training example once. In this work, we compared the new algorithm empirically with the most known existing ensemble methods for online learning. The experiments carried out show that the proposed algorithm frequently reaches higher levels of accuracy in the selected data sets.

Palabras clave : Data stream; Classifier ensemble; Concept drift.

        · resumen en Español     · texto en Español     · Español ( pdf )