Robust Outlier Detection and Low-Latency Concept Drift Adaptation for Data Stream Regression: A Dual-Channel Architecture

Outlier detection and concept drift detection represent two challenges in data analysis. Most studies address these issues separately. However, joint detection mechanisms in regression remain underexplored, where the continuous nature of output spaces makes distinguishing drifts from outliers inherently challenging. To address this, we propose a novel robust regression framework for joint outlier and concept drift detection. Specifically, we introduce a dual-channel decision process that orchestrates prediction residuals into two coupled logic flows: a rapid response channel for filtering point outliers and a deep analysis channel for diagnosing drifts. We further develop the Exponentially Weighted Moving Absolute Deviation with Distinguishable Types (EWMAD-DT) detector to autonomously differentiate between abrupt and incremental drifts via dynamic thresholding. Comprehensive experiments on both synthetic and real-world datasets demonstrate that our unified framework, enhanced by EWMAD-DT, exhibits superior detection performance even when point outliers and concept drifts coexist.

💡 Research Summary

In the rapidly evolving landscape of data stream analysis, maintaining the accuracy of regression models is a significant challenge due to the continuous influx of non-stationary data. This paper addresses a fundamental difficulty in data stream regression: the inherent ambiguity in distinguishing between point outliers and concept drifts. In a continuous output space, both phenomena manifest as increased prediction residuals, making it notoriously difficult to determine whether a spike in error is a transient anomaly or a fundamental shift in the underlying data distribution. To resolve this, the authors propose a novel, robust regression framework built upon a “Dual-Channel Decision Process.”

The core innovation of this research lies in its architectural split of prediction residuals into two coupled, specialized logic flows. The first, the “Rapid Response Channel,” is engineered for low-latency performance, specifically designed to identify and filter out point outliers almost instantaneously. This prevents the model from being corrupted by single erroneous data points. The second, the “Deep Analysis Channel,” operates on a different temporal scale, focusing on long-term pattern recognition to diagnose concept drifts. By decoupling these two processes, the framework can effectively manage the interference that often occurs when outliers and drifts coexist.

A pivotal contribution of this work is the development of the EWMAD-DT (Exponentially Weighted Moving Absolute Deviation with Distinguishable Types) detector. Unlike traditional detectors that primarily focus on the presence of change, EWMAD-DT utilizes exponential weighting and dynamic thresholding to characterize the nature of the drift itself. It possesses the autonomous capability to differentiate between “Abrupt Drifts,” which occur suddenly, and “Incremental Drifts,” which manifest as gradual changes over time. This granularity is crucial for adaptive learning, as it allows for more intelligent and efficient model retraining strategies.

The effectiveness of the proposed framework was rigorously evaluated using both synthetic and real-world datasets. The experimental results demonstrate that the dual-channel architecture, empowered by the EWMAD-DT algorithm, achieves superior detection performance, particularly in complex scenarios where outliers and concept drifts occur simultaneously. The framework exhibits high precision in identifying drifts while maintaining high recall in filtering outliers, even under heavy noise. This research provides a significant advancement for industries relying on real-time predictive analytics, such as financial fraud detection, industrial IoT monitoring, and autonomous systems, where robustness against both noise and distribution shifts is paramount for operational reliability.