NetBurst: Event-Centric Forecasting of Bursty, Intermittent Time Series

2025-10-25

AI Summary

Network telemetry time series—spanning service-, IP-, and subnet-level traffic—exhibit strong burstiness, prolonged inactive periods, and heavy-tailed distributions, fundamentally differing from conventional smooth, seasonal benchmarks; existing AI forecasting methods inadequately model these characteristics. To address this, we propose an event-centric modeling paradigm that, for the first time, integrates Mandelbrot’s heavy-tailed burstiness theory into deep learning: burst intensity is quantized via quantile-based codebooks, and a dual autoregressive architecture jointly models burst timing and magnitude while preserving event-level interpretability. Evaluated on large-scale real-world network traces, our method reduces MASE error by 13–605× over state-of-the-art baselines (e.g., Chronos) and improves embedding clustering clarity by 5×. This advances accurate, interpretable forecasting for intermittent bursty sequences.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Planning, Routing, and Scheduling: Model-Based Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Web Mining and Content Analysis: Models for Web evolution
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Forecasting on widely used benchmark time series data (e.g., ETT, Electricity, Taxi, and Exchange Rate, etc.) has favored smooth, seasonal series, but network telemetry time series -- traffic measurements at service, IP, or subnet granularity -- are instead highly bursty and intermittent, with heavy-tailed bursts and highly variable inactive periods. These properties place the latter in the statistical regimes made famous and popularized more than 20 years ago by B.~Mandelbrot. Yet forecasting such time series with modern-day AI architectures remains underexplored. We introduce NetBurst, an event-centric framework that reformulates forecasting as predicting when bursts occur and how large they are, using quantile-based codebooks and dual autoregressors. Across large-scale sets of production network telemetry time series and compared to strong baselines, such as Chronos, NetBurst reduces Mean Average Scaled Error (MASE) by 13--605x on service-level time series while preserving burstiness and producing embeddings that cluster 5x more cleanly than Chronos. In effect, our work highlights the benefits that modern AI can reap from leveraging Mandelbrot's pioneering studies for forecasting in bursty, intermittent, and heavy-tailed regimes, where its operational value for high-stakes decision making is of paramount interest.

Problem

• Forecasting bursty intermittent network telemetry time series
• Reformulating prediction as burst timing and magnitude estimation
• Addressing heavy-tailed regimes with quantile-based autoregressive frameworks

Innovation

• Event-centric framework predicting burst timing and size
• Quantile-based codebooks with dual autoregressors
• Leveraging Mandelbrot's statistical regimes for bursty data