Fast Adapting Ensemble: A New Algorithm for Mining Data Streams with Concept Drift

Abstract

The treatment of large data streams in the presence of concept drifts is one of the main challenges in the field of data mining, particularly when the algorithms have to deal with concepts that disappear and then reappear. This paper presents a new algorithm, called Fast Adapting Ensemble (FAE), which adapts very quickly to both abrupt and gradual concept drifts, and has been specifically designed to deal with recurring concepts. FAE processes the learning examples in blocks of the same size, but it does not have to wait for the batch to be complete in order to adapt its base classification mechanism. FAE incorporates a drift detector to improve the handling of abrupt concept drifts and stores a set of inactive classifiers that represent old concepts, which are activated very quickly when these concepts reappear. We compare our new algorithm with various well-known learning algorithms, taking into account, common benchmark datasets. The experiments show promising results from the proposed algorithm (regarding accuracy and runtime), handling different types of concept drifts.

Untitled

“…En estos escenarios no se sabe que tipos de cambios están presentes por lo que es importante realizar experimentos con datos de distintos escenarios. Los conjuntos de datos reales seleccionados (ver Tabla 1) se han usado en diferentes estudios sobre aprendizaje a partir de flujos de datos no estacionarios [13,27].…”

“…Los cambios abruptos ocurren cuando la transición entre conceptos consecutivos es instantánea y los cambios graduales ocurren cuando el período de transición contiene cierto número de ejemplos. Otro tipo común de cambio es el recurrente, que ocurre cuando los conceptos pueden reaparecer [27].…”

A survey of active and passive concept drift handling methods

“…Each ensemble member is assigned a different size chunk, and once a change is detected, a new classifier will be constructed to replace the worst classifier in the ensemble. Díaz et al 73 proposed a new fast adaptive ensemble algorithm (FAE). FAE stores a set of inactive basic classifiers to represent old concepts that have been analyzed and then disappeared.…”

Continual Learning for Predictive Maintenance: Overview and Challenges

“…In this work, the authors propose a combination of ensembles to generate diversity in the performance to detect different types of distribution drifts. Several works have used ensembles to detect drifts, including those by [1,160,187,188].…”