National and Kapodistrian University of Athens
Στατιστική ανάλυση και πρόβλεψη της έκβασης τραυματιών με χρήση αποδοτικών αλγόριθμων εξόρυξης πληροφορίας και ποινικοποιημένων μεθόδων επιλογής μεταβλητών
Abstract
dc:descriptionIn all sciences experiments are conducted in order to examine and reveal something about a process or system. In this thesis are presented the results of a new approach for selecting the most important predictors of a phenomenon, a combination of recently developed methods of data mining and penalized likelihood to high-dimensional medical problems. The method is used to find factors associated more with the possibility of death in people with traumatic injuries. Initially, data mining methods are presented. Steps that are followed during the procedure are described, which involve setting goals and using appropriate efficient algorithms for finding clustering, correlation, and classification rules. In detail the algorithms used for that purpose are A-priori, CRT, C5.0, CHAID, GRI, K-means and Two-step. Also to ensure the quality of results it is important to proper prepare data and therefore are presented in detail all the approaches to missing values frequently appear in the collection of medical information. Following are described the methods of selecting variables using penalized likelihood estimators who act like an oracle procedure, in the sense of choosing the correct model when the normalization parameter is selected properly. One of the advantages of these methods is that they detect significant variables and estimate the regression coefficients simultaneously. Also the procedure for calculating the standard error using a type of sandwich, is presented. Moreover, this thesis includes data collection methodology, and type of variables included in the study. Additionally, the results of extensive analysis of missing data are examined, including missing values mechanism, appropriate transformations of variables and conclusions from graphical information. Finally, the results are presented with the selection of important variables that lead to death trauma patients, using the above mentioned new approaches. Also the effectiveness of these methods and comparisons of the evaluation results according to their predictive ability, are discussed.
Degree
thesis:*- Grantor dc:publisher
- National and Kapodistrian University of Athens
- Year dc:date
- 2010
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Theodoraki, Eleni-Maria
- Θεοδωράκη, Ελένη-Μαρία
Subjects
dc:subject × 12Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/25362