Aristotle University Of Thessaloniki (AUTH)
Ανθεκτική παλινδρόμηση με μαθηματικό προγραμματισμό
Abstract
dc:descriptionThe aim of the present thesis was the development of new robust regression estimators and the introduction of diagnostic measures for the identification of outliers in data sets whose variables deviate from the ideal normality assumptions. The basic target of the thesis was the presentation of the advantages of the mathematical programming technique through which the identification of new types of robust regression estimators is feasible which belong in the class of M, GM or in high breakdown point estimators as LTS, S and MM. The comparison of the different new robust estimators with the classical famous estimators of the robust literature indicated the utility of the new estimators and also the importance of the introduction of mathematical programming as a very useful technique in robust regression. More specifically, we developed a new robust covariance estimator. A faster solution of the robust M and GM estimators was succeeded, compared to the classical iterated method IRLS, with the insertion of restrictions in the mathematical programming models. We presented two new maximum likelihood robust estimators of scale for the high breakdown robust estimator PTS, where the number of the outliers is not required a priori. Furthermore, we introduced a new high breakdown robust estimator with bounded influence function. Finally, we developed a new support vector machine’s model in regression.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2010
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Avramidis, Antonios
- Αβραμίδης, Αντώνιος
Subjects
dc:subject × 18- Στατιστική
- Ανθεκτική παλινδρόμηση
- Μαθηματικός προγραμματισμός
- Μέθοδος ποινικοποίησης
- Μικτός ακέραιος προγραμματισμός
- Ανίχνευση ακραίων παρατηρήσεων
- Μηχανισμός διανυσμάτων υποστήριξης
- Statistics
- Robust regression
- Mathematical programming
- Penalty method
- Mixed integer programming
- Identifying outliers
- Support vector machines
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Computer and Information Sciences
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/23932