Summary: | 碩士 === 國立臺灣科技大學 === 電子工程系 === 90 === In recent years, many researchers focused on the research topic of generating rules from training instances, where the decision tree method is a well-known method among them. The decision tree method can generate useful rules from a set of training data, but the data in the database is not usually suitable for parting by a precise point. The fuzzy decision tree method can overcome the drawback and can generate fuzzy rules from training instances. In this thesis, we present a new method to estimate null values in relational database systems, where we consider the attributes appearing in the antecedent portions of the generated fuzzy rules have different weights, and we apply the weights of the attributes to derive the certainty factor (CF) value of each generated fuzzy rule for generating better fuzzy rules for estimating null values in relational database systems. Furthermore, we also present a method to derive the values of hypothetical certainty factor (HCF) nodes for constructing a complete fuzzy decision tree for generating better fuzzy rules for estimating null values in relational database systems.
In this thesis, we also present another new method for estimating null values in relational database systems based on genetic algorithms. We consider that the experts predefine the initial membership functions of the linguistic terms, and the proposed method tune the membership functions by using a genetic algorithm and generate fuzzy rules to get a higher average estimated accuracy rate. The proposed methods can get higher average estimated accuracy rates than the existing methods for estimating null values in relational database systems.
|