Fun-TSG: A Function-Driven Multivariate Time Series Generator with Variable-Level Anomaly Labeling

2026-04-14

AI Summary

Existing benchmarks for multivariate time series anomaly detection commonly lack fine-grained anomaly annotations, explicit modeling of inter-variable and temporal dependencies, and transparent generation mechanisms, thereby hindering reliable evaluation of interpretable, variable-level methods. This work proposes a function-driven generative framework that explicitly encodes dynamic inter-variable dependencies and temporal evolution through functional modeling, random dependency graph sampling, and user-defined equation systems. The framework enables precise, controllable injection of anomalies at both variable and timestamp granularity. For the first time, it achieves full transparency in the generation process, configurable dependency structures, and fine-grained, customizable anomaly labeling. Consequently, it facilitates the construction of diverse, interpretable, and reproducible benchmark scenarios, substantially enhancing the fine-grained evaluation of existing anomaly detection models at the variable level.

Technology Category

• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Web Mining and Content Analysis: Web data generation and simulation
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Reliable evaluation of anomaly detection methods in multivariate time series remains an open challenge, largely due to the limitations of existing benchmark datasets. Current resources often lack fine-grained anomaly annotations, do not provide explicit intervariable and temporal dependencies, and offer little insight into the underlying generative mechanisms. These shortcomings hinder the development and rigorous comparison of detection models, especially those targeting interpretable and variable-specific outputs. To address this gap, we introduce Fun-TSG, a fully customizable time series generator designed to support high-quality evaluation of anomaly detection systems. Our tool enables both fully automated generation, based on randomly sampled dependency structures and anomaly types, and manual generation through user-defined equations and anomaly configurations. In both cases, it provides full transparency over the data generation process, including access to ground-truth anomaly labels at the variable and timestamp levels. Fun-TSG supports the creation of diverse, interpretable, and reproducible benchmarking scenarios, enabling fine-grained performance analysis for both classical and modern anomaly detection models.

Problem

• multivariate time series
• anomaly detection
• benchmark dataset
• anomaly labeling
• generative mechanism

Innovation

• function-driven generation
• multivariate time series
• variable-level anomaly labeling
• interpretable benchmarking
• customizable synthetic data