Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LLM-Augmented Changepoint Detection: A Framework for Ensemble Detection and Automated Explanation

Fabian Lukassen, Christoph Weisser, Michael Schlee, Manish Kumar, Anton Thielmann, Benjamin Saefken, Thomas Kneib

TL;DR

The paper tackles the problem of interpreting changepoints by jointly improving detection robustness and providing automatic, context-aware explanations. It introduces an LLM-Augmented Changepoint Detection framework that combines an ensemble of ten detectors with a transparent spatial-clustering and voting mechanism, plus a dual-mode LLM explanation pipeline (Standard and RAG) to ground narratives in public knowledge or private documents. Key contributions include a practical ensemble approach with interpretable aggregation, an automatic method-selection scheme guided by data profiling, and a RAG-enabled explanation workflow that preserves privacy while enabling domain-specific attributions. The framework is validated on seven diverse datasets, showing superior detection performance and higher-quality explanations, and a synthetic RAG example demonstrates grounding when private documents exist and calibrated uncertainty when they do not, highlighting practical implications for finance, politics, and environment domains.

Abstract

This paper introduces a novel changepoint detection framework that combines ensemble statistical methods with Large Language Models (LLMs) to enhance both detection accuracy and the interpretability of regime changes in time series data. Two critical limitations in the field are addressed. First, individual detection methods exhibit complementary strengths and weaknesses depending on data characteristics, making method selection non-trivial and prone to suboptimal results. Second, automated, contextual explanations for detected changes are largely absent. The proposed ensemble method aggregates results from ten distinct changepoint detection algorithms, achieving superior performance and robustness compared to individual methods. Additionally, an LLM-powered explanation pipeline automatically generates contextual narratives, linking detected changepoints to potential real-world historical events. For private or domain-specific data, a Retrieval-Augmented Generation (RAG) solution enables explanations grounded in user-provided documents. The open source Python framework demonstrates practical utility in diverse domains, including finance, political science, and environmental science, transforming raw statistical output into actionable insights for analysts and decision-makers.

LLM-Augmented Changepoint Detection: A Framework for Ensemble Detection and Automated Explanation

TL;DR

The paper tackles the problem of interpreting changepoints by jointly improving detection robustness and providing automatic, context-aware explanations. It introduces an LLM-Augmented Changepoint Detection framework that combines an ensemble of ten detectors with a transparent spatial-clustering and voting mechanism, plus a dual-mode LLM explanation pipeline (Standard and RAG) to ground narratives in public knowledge or private documents. Key contributions include a practical ensemble approach with interpretable aggregation, an automatic method-selection scheme guided by data profiling, and a RAG-enabled explanation workflow that preserves privacy while enabling domain-specific attributions. The framework is validated on seven diverse datasets, showing superior detection performance and higher-quality explanations, and a synthetic RAG example demonstrates grounding when private documents exist and calibrated uncertainty when they do not, highlighting practical implications for finance, politics, and environment domains.

Abstract

This paper introduces a novel changepoint detection framework that combines ensemble statistical methods with Large Language Models (LLMs) to enhance both detection accuracy and the interpretability of regime changes in time series data. Two critical limitations in the field are addressed. First, individual detection methods exhibit complementary strengths and weaknesses depending on data characteristics, making method selection non-trivial and prone to suboptimal results. Second, automated, contextual explanations for detected changes are largely absent. The proposed ensemble method aggregates results from ten distinct changepoint detection algorithms, achieving superior performance and robustness compared to individual methods. Additionally, an LLM-powered explanation pipeline automatically generates contextual narratives, linking detected changepoints to potential real-world historical events. For private or domain-specific data, a Retrieval-Augmented Generation (RAG) solution enables explanations grounded in user-provided documents. The open source Python framework demonstrates practical utility in diverse domains, including finance, political science, and environmental science, transforming raw statistical output into actionable insights for analysts and decision-makers.
Paper Structure (90 sections, 7 equations, 4 figures, 12 tables, 2 algorithms)

This paper contains 90 sections, 7 equations, 4 figures, 12 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Ensemble Changepoint Detection
  4. Event Attribution and Automated Explanation
  5. LLMs for Time Series Analysis
  6. Retrieval-Augmented Generation (RAG)
  7. Research Gaps and Our Contribution
  8. Problem Statement
  9. Methodology
  10. Changepoint Detection Strategies
  11. Individual Method Selection
  12. Automatic Method Selection
  13. Data Profiling
  14. Method Filtering and Scoring
  15. Ensemble Method
  16. ...and 75 more sections

Figures (4)

  • Figure 1: Structural break analysis in financial time series. Detected changepoints (red dashed lines) correspond to major real-world events. This illustrates our framework's core challenge: automatically linking statistical anomalies to their historical causes.
  • Figure 2: Complete workflow of the LLM-augmented changepoint detection framework. The system processes time series data through three main stages: (1) changepoint detection (2) optional RAG integration for private data, retrieving relevant documents using hybrid semantic-temporal search; and (3) LLM-based causal attribution, which identifies plausible events underlying the detected changepoints.
  • Figure 3: Interactive web interface: Step-by-step workflow with data upload via drag-and-drop or sample data, followed by column selection, analysis configuration, and results viewing.
  • Figure 4: Interactive web interface: Results visualization showing detected structural breaks with labeled change directions, time range filtering, and export functionality.