A Representation Learning Approach to Feature Drift Detection in Wireless Networks

2025-05-15

AI Summary

To address the challenge of timely detecting AI model performance degradation caused by feature distribution drift in wireless networks, this paper proposes ALERT—a three-stage framework integrating representation learning (via MLP), statistical testing (KS/PSI), and application-aware utility evaluation to enable automated drift detection and trigger retraining. Its key innovation lies in being the first to jointly incorporate representation learning and utility-driven evaluation into a unified drift detection paradigm, along with a novel utility function that jointly balances detection sensitivity and operational impact—overcoming the limitation of conventional methods that solely quantify distributional divergence. Evaluated on two real-world wireless scenarios—fingerprint-based indoor localization and link anomaly detection—ALERT consistently outperforms ten state-of-the-art baselines, achieving up to a 32% improvement in detection accuracy and significantly reducing false-positive retraining triggers.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Search and Optimization: Distributed Search
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

AI is foreseen to be a centerpiece in next generation wireless networks enabling enabling ubiquitous communication as well as new services. However, in real deployment, feature distribution changes may degrade the performance of AI models and lead to undesired behaviors. To counter for undetected model degradation, we propose ALERT; a method that can detect feature distribution changes and trigger model re-training that works well on two wireless network use cases: wireless fingerprinting and link anomaly detection. ALERT includes three components: representation learning, statistical testing and utility assessment. We rely on MLP for designing the representation learning component, on Kolmogorov-Smirnov and Population Stability Index tests for designing the statistical testing and a new function for utility assessment. We show the superiority of the proposed method against ten standard drift detection methods available in the literature on two wireless network use cases.

Problem

• Detects feature distribution changes in wireless networks
• Triggers model re-training to prevent AI performance degradation
• Evaluates effectiveness on wireless fingerprinting and anomaly detection

Innovation

• Uses representation learning for feature drift detection
• Applies statistical testing with KS and PSI methods
• Introduces new utility assessment function for re-training