PyOD 2: A Python Library for Outlier Detection with LLM-powered Model Selection
Sihan Chen, Zhuangzhuang Qian, Wingchun Siu, Xingcan Hu, Jiaqi Li, S. J. Li, Yuehan Qin, Tiankai Yang, Zhuo Xiao, Wanghao Ye, Yichi Zhang, Yushun Dong, Yue Zhao · 2025
Outlier detection (OD), or anomaly detection, is essential in data analysis and machine learning. The Python Outlier Detection (PyOD) library is the most popular open-source tool for OD-with over 8,500 GitHub stars and 25 million downloads-but it has three limitations: (1) limited modern deep learning algorithms, (2) fragmented PyTorch and TensorFlow implementations, and (3) no automated model selection, which challenges non-experts. To overcome these issues, we introduce PyOD Version 2 (øurs), which integrates 10 state-of-the-art deep learning models into a unified PyTorch framework and adds an LLM-based pipeline for automated model selection. These enhancements streamline OD workflows, offer access to 45 algorithms, and deliver robust performance across various datasets. In this paper, we demonstrate how øurs simplifies the deployment and automation of OD models, setting a new standard in research and industry. øurs is available at https://github.com/yzhao062/pyod.