الفهرس 
Table of ContentOnly 14 pages are availabe for public view
 |  | from | 200 |  |  |
| | | from | 200 | | |
Abstract
The concern about knowledge representation and information systems, especially real valued, uncertain and ambiguous dataset, has put forward useful extensions of classical set theory. The knowledge base systems are suffering from: preparing regular attributes (features) to be credible fuzzy ones with high ranking measures, reducing the features dimensionality, finding accurate and efficient fuzzy rules, calculating the credibility level of the whole system and enhancing the concluded rules to estimate the unseen objects. Based on the credibility theory, this thesis introduces mining of an efficient set of fuzzy rules that are inferred by a hybrid model of Soft Computing techniques. In order to make efficient and completely automated knowledge based system (without any biased expert interfering), these problems need good machine learning algorithms to design a complete hybrid system in the field of the data under consideration.
This thesis proposes completely automated hybrid models of rule based system as an example for knowledge based systems. The proposed models are transformation hybrid models. In the first hybrid model, the Self Organized Features Maps (SOFM) uses its clustering capabilities to design the fuzzy variables. Evolutionary algorithms like Parallel Genetic Algorithms (PGAs), as a search mechanism, find the accurate and efficient fuzzy (if-then) rule set that represent that data. The set of fuzzy (if-then) rules created are tested using a test data set to decide its credibility measure. The second hybrid is a fuzzy system using credibility measure. It extends the first model by a module for ranking the fuzzy variables, generated from Self Organized Features Maps, using the credibility Inversion theory. The credible fuzzy variables are passed to the FRANTIC-SRL (Fuzzy Rules from ANT-Inspired Computation – Simultaneous Rule Learning)for designing the fuzzy (if-then)rules. The credibility measure is invoked again to calculate the confidence level of the classes attribute (the system output after applying the unseen instances). The system reliability depends on the credibility of the classes attribute.
In the third proposed system, hybrid fuzzy rough system enhanced by fuzzy cellular automata, the fuzzy variables are generated and ranked in the same way like the previous proposed fuzzy systems. The Fuzzy Rough Attribute Reduction (FRAR) algorithm reduces the features basing on measuring the dependency membership degree between the fuzzy variables and the training data set to produce the reduct. This system generates the fuzzy rough (if-then) rules using the data summarizing technique of the rough set theory. Moreover, the system uses the credibility inversion theory on the classes attribute (system output) to calculate the reliability of the whole system. The fuzzy cellular automata (FCA) are parallel systems. FCA takes the fuzzy and fuzzy rough (if-then) rules (system equation) as an initial state. Moreover, it reproduces it in the time sequence according to fuzzy n4V1 nonsatble update rule to get an estimate of the system in the future. This estimate helps the experts to avoid crises or to push the system in certain direction to increase the system efficiency. The three proposed systems help decision makers to extract valuable knowledge with acceptable degrees of truth. In contrast with other approaches, a comparison with approaches published in literature is illustrated. The comparative study shows the efficiency of the proposed hybrid models using the error rate as the accuracy measure. The details and limitations of the new approaches are discussed and the future works are suggested.