Abstract
dc:descriptionIn this thesis a generic methodology for the automated generation of fuzzy models is presented. The methodology includes three stages: (i) crisp model extraction, (ii) fuzzy model creation, and (iii) parameter optimization. In the first stage, a crisp model is created; depending on the approach, which will be employed for the crisp model creation, the methodology can be knowledge-based, if the initial crisp model is defined by experts, or data-driven, if the initial crisp model is mined from the available data. In the second stage, this crisp model is transformed to the corresponding fuzzy model; several new parameters are introduced due to the fuzzification of the decision boundaries. Finally (in the third stage), optimization is performed in order to optimally define all the parameters entering the fuzzy model. The methodology is generic, thus different realizations can be created. The realizations created are the crisp approach, which consists only from the first stage of the methodology, the knowledge based approach, where the initial crisp set of rules is defined by experts, and the data-driven approach, where data mining techniques are employed in order to generate the initial crisp set of rules directly from the data. In some cases weights are included in the fuzzy model. The methodology is applied to the cardiovascular diseases domain. More specifically, the methodology is employed in order to automatically generate fuzzy models that deal with the problem of arrhythmic beat classification in ECG, the ischemic beat classification in ECG and the coronary artery disease diagnosis. In all cases, the fuzzy models present comparable or higher accuracy than other approaches presented in the literature. In addition, a large number of benchmark classification datasets is used for the evaluation and the reported results indicate high classification accuracy. The methodology presents several advantages and novelties: it is generic since it is not based on a specific technique for crisp model generation; expert knowledge or any rule-mining technique can be adapted to generate the crisp model. This offers flexibility, since the methodology does not depend on the domain of application. This is advantageous since it can integrate state-of-the-art rule mining methods, as well as future developments or even hybrid approaches, combining expert knowledge with the mined knowledge. Also, different approaches concerning the elements of the fuzzy model and alternative optimization techniques can be integrated. Another major advantage of the methodology is that the transformation of the crisp model to the respective fuzzy one, is straightforward, ensuring in this way the fully automated nature of the methodology. In the case of the data-driven approach, the generation of fuzzy models, based on the fuzzification and optimization of a set of rules extracted from the decision tree, instead of a fuzzy decision tree (widely proposed in the literature), is a novel feature which offers more flexibility and adaptation ability to a specific dataset, while keeping the complexity of the decision making process the same. Finally, the introduction of class weights is a novel feature which allows the fuzzy model to be more flexible and adaptable.
Degree
thesis:*- Grantor dc:publisher
- University of Ioannina
- Year dc:date
- 2008
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Tsipouras, Markos
- Τσίπουρας, Μάρκος
Subjects
dc:subject × 17- Αυτόματη διάγνωση
- Καρδιαγγειακές παθήσεις
- Ασαφής λογική
- Γνωσιακά μοντέλα
- Ασαφή μοντέλα
- Δέντρα αποφάσεων
- Automated diagnosis
- Intelligent methods
- Cardiovascular disease
- Fuzzy logic
- Knowledge based models
- Fuzzy models, Automated creation of
- Decision trees
- 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/17723