.png){kind=icon}
Return
Adaptive XGBoost for Data Stream Regression
.png){kind=icon}
OA
AI
DOI：10.1155/int/1759600.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This work introduces Lean Adaptive XGBoost for Regression (LAX-Reg), a novel, high-efficiency algorithm designed for intelligent processing of nonstationary data streams. The rising volume and velocity of data generated by connected systems necessitate models that can adapt in real time to concept drift. While data stream classification has been extensively studied, data stream regression, particularly using boosting-based methods, remains comparatively underexplored. LAX-Reg builds on XGBoost while avoiding the alternating-model paradigm commonly adopted in existing stream adaptations, which can lead to abrupt performance degradation during model replacement and increased training overhead. Instead, it maintains a single, continuously updated ensemble with bounded complexity through dynamic tree management: outdated trees are selectively removed, remaining trees are updated, and new trees are added using either a FIFO strategy or a Target strategy driven by individual ADWIN drift detectors. The proposed approach is evaluated using prequential evaluation on nine real and synthetic data streams, considering mean squared error (MSE) alongside execution times and memory consumption and compared against five state-of-the-art baselines, including ARF-Reg and AFXGB. Experimental results show that LAX-Reg variants achieve the best average predictive rankings and belong to the leading statistical group according to the Nemenyi post hoc test, while delivering substantial efficiency gains: up to 112 × faster execution and 456 × lower memory usage compared with ARF-Reg. These results highlight LAX-Reg as a competitive and lightweight solution for adaptive regression in data streams, while also motivating future work on alternative drift detectors, recurring drift scenarios, and the incorporation of temporal features.
Keywords:
data stream
decision trees
nonstationary time series
regression
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
3.7
Papers:
3.1K
Citations:
8.1K
Organization
Cited Papers
No cited papers available