ADAL-R: Adaptive Drift-Aware Active Learning for Label-Efficient Streaming Regression

Learning from continuous data streams under limited labeling budgets remains a major challenge in modern machine learning applications. While active learning reduces annotation cost, most existing approaches are designed for static settings and fail to adapt to streaming regression with evolving data distributions. This limitation is critical in real-world scenarios where concept drift is unavoidable and labeling is expensive. In this paper, we propose ADAL-R, an adaptive drift-aware active learning framework for streaming regression that integrates uncertainty estimation, density-aware sampling, drift detection, and adaptive query thresholding. The framework dynamically adjusts its labeling strategy based on predictive uncertainty and environmental changes. Experimental results on synthetic and real-world datasets demonstrate that ADAL-R achieves strong predictive performance while querying fewer than 10\% of labels. These findings highlight the effectiveness of combining drift awareness with adaptive active learning for efficient learning systems.