PyOD is a Python library for anomaly detection that supports tabular, time series, graph, text, image, and audio data.
The library addresses the challenge of detecting anomalies across diverse data modalities by providing a unified interface to multiple detection algorithms. It includes over 60 detectors spanning classical statistical methods, machine learning approaches, and deep learning models. The tool orchestrates these detectors through an ADEngine that uses benchmark results to route requests to the most appropriate algorithm for a given problem, eliminating the need to manually select among competing approaches.
Teams building fraud detection systems, data quality pipelines, or out-of-distribution monitoring should consider PyOD if they work with heterogeneous data types and want a single library rather than maintaining separate tools for tabular versus time series versus image anomalies. The project supports both traditional machine learning workflows through its standard fit and predict API, and integration with AI agents through an od-expert skill for Claude Code and Codex, plus MCP-compatible tools for structured agent queries. The library's detector selection is grounded in results from multiple benchmarks covering tabular data, time series, and NLP domains.
The project maintains active continuous integration with automated testing across its codebase. Development activity shows consistent engagement with code quality monitoring and maintainability tracking. The tool receives substantial real-world usage, reflected in its download volume and the breadth of detector implementations it sustains across multiple data modalities.
