Multiple outlier detection revisited

The discrimination power of the classical or/and robust diagnostics for the 34 (real and simulated) regression data sets with multiple outliers is compared. These diagnostics, presented in the uniform way for the large number of objects, are then used in the pattern recognition approaches (PLS, Neural Networks, Rough Set Theory) to estimate the joint discrimination power of the classical or/and robust diagnostics and to construct models (or logical rules), allowing identification of outliers in the new data sets.