pyod
Python library providing 60+ anomaly detection algorithms for tabular, graph, image, and text insurance claim data.
Detect suspicious claims, billing outliers, and network fraud using 60+ benchmarked algorithms.
PyOD is a comprehensive Python library designed for detecting outliers across tabular metrics, claim text, image evidence, and relational policy graphs. Insurance data science teams can use it to identify fraudulent claims, billing anomalies, and rare loss patterns. It provides a classic scikit-learn API alongside automated engine routing and AI agent integration.
What it is
PyOD organizes over 60 detection algorithms under a single standardized interface. Its ADEngine orchestrates automated detector selection, model evaluation, and result interpretation across multi-modal inputs. The library includes native skills for Claude Code and Codex, plus a Model Context Protocol (MCP) server for autonomous AI investigation.
Who it's for
Insurance data scientists, fraud analytics teams, and actuary engineers needing a production-ready toolkit to catch unusual claims should use PyOD. It is ideal for teams comfortable writing Python scripts or integrating LLM workflows. It is not suitable for non-technical adjusters seeking an out-of-the-box visual UI without writing code.
Key features
60+ Multi-Modal Detectors
Covers tabular data, time series, graphs, image evidence, and unstructured text within one consistent API.
ADEngine Orchestration
Automatically profiles datasets, selects optimal detection algorithms, and explains anomaly findings.
Agentic AI Integration
Provides the od-expert skill and an MCP server so AI assistants like Claude can run end-to-end investigations.
SUOD Parallel Training
Uses SUOD and Numba JIT compilation to accelerate model fitting and scoring across large claim volumes.
Up and running in 3 steps
Our honest verdict
What's great
- Supports tabular, text, image, and graph data in a single standardized Python API.
- Offers built-in agent skill integration for Claude Code and MCP-compatible AI assistants.
- Uses SUOD and Numba acceleration to handle large datasets fast.
- Provides benchmark-backed model selection via ADEngine to reduce guesswork.
Where it falls short
- Requires writing Python code or setting up LLM agent environments to operate.
- Lacks an out-of-the-box graphical user interface for non-technical claim handlers.
- Compute costs scale up when training complex deep learning detectors on heavy image or graph data.
- Deep learning and multi-modal extras require managing heavy secondary dependencies.
Pricing & license
PyOD is completely free and open-source under the BSD-2-Clause license. There are no paid commercial editions mentioned; costs are limited strictly to your own compute infrastructure and API fees if using external LLM agents.
Open remote AI-training roles
880 open remote AI-training roles, updated November 15, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
How does PyOD assist in insurance fraud detection?
PyOD identifies unusual patterns across claim amounts, medical code combinations, policy networks, and submitted photo evidence using over 60 specialized algorithms.
Can non-programmers use PyOD directly?
No, PyOD is a Python software library that requires programming experience or an AI agent framework like Claude Code to run.
What data formats are supported by PyOD?
It processes tabular databases, claim descriptions in text, time series billing logs, claim network graphs, and image files.
