AIEurolens: Explainable AI framework for drift detection applied to 5G time-series data

Time series forecasting plays a critical role in enabling machine learning–based resource management solutions for optimizing fifth generation (5G) and emerging (sixth generation) 6G networks. However, forecasting accuracy is often compromised by concept drift, which refers to shifts in data distributions over time. Existing drift detection approaches suffer from key limitations, including heightened sensitivity to noise, reliance on assumptions of static data properties, and inadequate handling of multivariate temporal dependencies, hence hindering Machine Learning (ML) model developers from diagnosing the root causes of time series model degradation. To address these challenges, this paper introduces AIEuroLens, an explainable framework that (1) detects drift by identifying significant shifts in temporal feature correlations, and (2) links these shifts to specific model deficiencies, such as suboptimal training procedures or architectural misconfigurations. This enables developers to distinguish between inherently unpredictable data dynamics and f laws that can be corrected through model improvements. Evaluated on real-world 5G network data, AIEuroLens demonstrates accurate and timely drift detection while also informing effective mitigation strategies. As part of ongoing research, this work contributes to the development of self-correcting forecasting systems for zero-touch 5G network management, helping to bridge the gap between drift detection and actionable model adaptation.