Explainable Heterogeneous Anomaly Detection in Financial Networks via Adaptive Expert Routing

2025-10-19

AI Summary

Financial anomalies exhibit heterogeneous mechanisms—including price shocks, liquidity freezes, contagion cascades, and institutional shifts—yet existing detection methods rely on static graph structures and uniform scoring, lacking mechanism identification capability and interpretability, thus impeding precise regulatory intervention. This paper proposes the first adaptive, interpretable detection framework tailored to heterogeneous anomaly mechanisms. It introduces mechanism-specific expert routing and a two-level attribution architecture to embed interpretability intrinsically; integrates stress-modulated dynamic graph learning with structural priors for unsupervised mechanism identification and temporal tracking; and models multi-scale spatiotemporal dependencies via BiLSTM–self-attention and cross-modal attention, while learning dynamic graph structures through neural multi-source interpolation. Evaluated on 100 U.S. stocks, the framework achieves a 92.3% detection rate for major events, with an average early warning lead time of 3.8 days—outperforming the best baseline by 30.8 percentage points—and demonstrates strong mechanistic traceability and evolutionary path characterization, as validated on the SVB case.

Technology Category

• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Machine Learning: Graph-based Machine Learning
• Multiagent Systems: Mechanism Design

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Web Mining and Content Analysis: Bridging structured and unstructured data
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Financial anomalies exhibit heterogeneous mechanisms (price shocks, liquidity freezes, contagion cascades, regime shifts), but existing detectors treat all anomalies uniformly, producing scalar scores without revealing which mechanism is failing, where risks concentrate, or how to intervene. This opacity prevents targeted regulatory responses. Three unsolved challenges persist: (1) static graph structures cannot adapt when market correlations shift during regime changes; (2) uniform detection mechanisms miss type-specific signatures across multiple temporal scales while failing to integrate individual behaviors with network contagion; (3) black-box outputs provide no actionable guidance on anomaly mechanisms or their temporal evolution. We address these via adaptive graph learning with specialized expert networks that provide built-in interpretability. Our framework captures multi-scale temporal dependencies through BiLSTM with self-attention, fuses temporal and spatial information via cross-modal attention, learns dynamic graphs through neural multi-source interpolation, adaptively balances learned dynamics with structural priors via stress-modulated fusion, routes anomalies to four mechanism-specific experts, and produces dual-level interpretable attributions. Critically, interpretability is embedded architecturally rather than applied post-hoc. On 100 US equities (2017-2024), we achieve 92.3% detection of 13 major events with 3.8-day lead time, outperforming best baseline by 30.8pp. Silicon Valley Bank case study demonstrates anomaly evolution tracking: Price-Shock expert weight rose to 0.39 (33% above baseline 0.29) during closure, peaking at 0.48 (66% above baseline) one week later, revealing automatic temporal mechanism identification without labeled supervision.

Problem

• Detecting heterogeneous financial anomalies with mechanism-specific interpretability
• Adapting to dynamic market correlations and multi-scale temporal dependencies
• Providing actionable anomaly attributions without post-hoc interpretation methods

Innovation

• Adaptive expert networks with specialized routing
• Multi-scale temporal dependencies via BiLSTM attention
• Dynamic graph learning with interpretable architectural embeddings