A0993
Title: Outlier detection in robust regression via chance-constrained programming
Authors: Hao Zhang - University of Arizona (United States) [presenting]
Abstract: Outlier detection is a critical step in data analysis to identify heterogeneous points in data. For high dimensional and extremely noisy data, many challenges are posed by outlier points, such as estimating the number of outliers, providing probabilistic confidence statements on identified outliers, fitting robust models against outliers, and achieving high breakdown points with a guarantee. To address these issues, we propose a chance-constrained outlier detection (CCOD) model that integrates robust regression and outlier diagnostics in one unified methodology. Theoretically, we prove that the new method can achieve a high breakdown point. To tackle the nonconvex computational problem, we propose a tractable and scalable convex approximation. Numerical results show that our CCOD model outperforms the state-of-art methodologies in terms of estimation accuracy, robustness, and computational time.
