Active feature acquisition on data streams under feature drift

• DOI: 10.1007/s12243-020-00775-2
• Publication Date: 2020-07-08T12:03:44Z
• Authors: Christian Beyer; Maik Büttner; Vishnu Unnikrishnan; Miro Schleicher; Eirini Ntoutsi; Myra Spiliopoulou
• Keywords: Feature (linguistics); Concept Drift; Baseline (sea)

Abstract Traditional active learning tries to identify instances for which the acquisition of label increases model performance under budget constraints. Less research has been devoted task actively acquiring feature values, whereupon both instance and must be selected intelligently even less a scenario where arrive in stream with drift. We propose an strategy data streams drift, as well evaluation framework. also implement baseline that chooses features randomly compare random approach against eight different methods we can acquire at most one time per all are considered cost same. Our initial experiments on 9 sets, 7 degrees missing 8 budgets show our developed outperform sets have comparable remaining two.