Cellwise and Casewise Robust Covariance in High Dimensions
Researchers have developed a new method called cellRCov to address the sensitivity of covariance matrices to outliers in high-dimensional data. This method can simultaneously handle casewise outliers, cellwise outliers, and missing data, overcoming limitations of previous robust estimators that were only feasible up to 20 dimensions. The cellRCov method leverages robust PCA and ridge-type regularization, and its theoretical properties, including consistency and asymptotic normality, have been established. Simulations show its superior performance in contaminated and missing data scenarios, with practical applications in anomaly detection and robust canonical correlation analysis. AI
IMPACT Introduces a novel statistical technique for handling outliers, potentially improving the robustness of machine learning models that rely on covariance estimation.