Abstract:Detecting concept drift in high-speed data streams remains challenging, particularly when models must operate on unlabeled data and avoid false alarms caused by benign shifts. While disagreement-based uncertainty has shown promise in neural networks, its adaptation to ensembles of incremental decision trees (IDTs) remains largely unexplored. We investigate this approach by constructing batch-specific disagreement measures via label flipping in ensemble members and evaluating their effectiveness for drift detection in tabular data streams. Our experiments show that, although this method performs well in ensembles of multi-layer perceptrons (MLPs), it consistently underperforms loss-based detectors when applied to IDTs. We attribute this behavior to the intrinsic rigidity of IDTs: learning primarily through structural expansion, with limited parameter adaptation, restricts model plasticity and prevents disagreement from reliably reflecting learning potential. Recent work on restructuring IDTs using their intrinsic decomposition into non-overlapping rules offers a promising direction for improving adaptability.
| Comments: | Published as a conference paper at CAO Workshop at ICLR 2026 |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2605.12803 [cs.LG] |
| (or arXiv:2605.12803v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2605.12803 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Afonso Lourenço [view email]
[v1]
Tue, 12 May 2026 22:53:03 UTC (467 KB)