Supplementary Material Automatic Unsupervised Outlier Model Selection

Model set Mis composed by pairing outlier detection algorithms to distinct hyperparameter choices. Table 2 provides a comprehensive description of models, including 302 unique models composed by 8 popular outlier detection (OD) algorithms. All models and parameters are based on the Python Outlier Detection Toolbox (PyOD)5. B.1 Complete List of Meta-features We summarize the meta-features used by METAOD in Table 3. When applicable, we provide the formula for computing the meta-feature(s) and corresponding variants.