Outlier detection by means of robust regression estimators for use in engineering science

Abstrakt

The paper compares the ability of different robust regression estimators to detect and classify outliers. Estimators with a high breakdown point are compared and conclusions are drawn for real engineering applications.

The least trimmed squares estimator is recommended for heavily contaminated data sets with outliers with a complicated multivariate structure.

Klíčová slova

• Robust statistics
• outlier detection
• linear regression
• breakdown point
• engineering science