Detecting outliers using COPOD
COPOD is an exciting algorithm based on a paper published in September 2020, which you can read here: https://arxiv.org/abs/2009.09463.
The PyOD library offers many algorithms based on the latest research papers, which can be broken down into linear models, proximity-based models, probabilistic models, ensembles, and neural networks.
COPOD falls under probabilistic models and is labeled as a parameter-free algorithm. The only parameter it takes is the contamination factor, which defaults to 0.1. The COPOD algorithm is inspired by statistical methods, making it a fast and highly interpretable model. The algorithm is based on copula, a function generally used to model dependence between independent random variables that are not necessarily normally distributed. In time series forecasting, copulas have been used in univariate and multivariate forecasting, which is popular in financial risk modeling. The term copula stems from the copula function joining...
