Hierarchical Reduced-space Drift Detection Framework for Multivariate Supervised Data Streams

In a streaming environment, the characteristics of data themselves and their relationship with labels are likely to experience changes as time goes on. Most drift detection methods for supervised streams performance-based, that is, they detect only after classication accuracy deteriorates. This may not be sufcient in many application areas where reason behind is also important. Another category detectors distribution-based detectors. Although can some drifts within input space, affecting labelling mechanism cannot identied. Furthermore, little work available on high-dimensional streams. this paper we propose an advanced Hierarchical Reduced-space Drift Detection Framework Supervised Data Streams (HRDS) which captures regardless effects performance. framework suggests monitoring both marginal class-conditional distributions lower-dimensional space specically relevant assigned task. Experimental comparisons have demonstrated proposed HRDS achieves high-quality performance streams, but outperforms its competitors terms recall, precision F-measure across wide range different concept types including subtle drifts.