Thesis (Ph.D.) - Cairo University - Institute of Statistical Studies and Research - Department of Statistics and Econometrics

We consider the problem of handling outliers in classification models. Many real data sets contain outliers and these outliers may have bad effects on the estimation of parameters ofclassification models, also they affect predictions, classification errors and conclusions drawn from such models. The current research handles the problem of outliers presenting robust estimation methods in logistic and discriminant analysis. We also propose a new robust estimation method in logistic regression that depends on using a loss function which is to be trimmed on extreme outliers based on lemma derived by the researcher. Simulation studies have been conducted to compare between unpenalized and penalized logistic methods. Also, Simulation studies have been conducted to compare between two robust multivariate estimators using covering region and Fisher discriminant methods. Finally, three real-life examples have been analyzed to confirm the results of simulation studies

Issued also as CD