An Online Incremental Adaptation Mechanism to Subdue the Effect of Drift in Streaming Data

Abstract

Concept drift detection and adaptation is one of the crucial components of a resilient machine learning pipeline in production. The Adaboost is an ensemble approach that incorporates incremental learning, that is widely used for concept drift adaptation in streaming data. It is generally combined with other methods such as ARF classifiers and Bagging Classifiers. This study presents a collection of online incremental learning algorithms for streaming data to adjust machine learning categorization when there is concept drift. Better results are obtained on the Australian power dataset, demonstrating the efficacy of our approach in comparison to the current benchmark.