الفهرس 
Table of contentsOnly 14 pages are availabe for public view
 |  | from | 124 |  |  |
| | | from | 124 | | |
Abstract
In this thesis, Granular Computing (GrC) is utilized as a renew methodology for KDD for classification. The main model is Rough Mereology based GrC model, the other models based on hybridization of Rough Sets Theory (RST) and different data mining techniques, such as Artificial Neural Networks (ANNs) and Genetic Algorithm (GA )in the shape of GrC to gain an effective KDD model having the capabilities to classify medical datasets with binary or multi-value decisions attribute and extracting a set of treatment decision rules that help in the process of diseases diagnosing.The experiments in this thesis using two kinds of diseases, which are Hepatitis C Virus (HCV) and Coronary Artery Disease (CAD), were used as case studies. At first, Rough Sets (RS) based GrC model is proposed. The first phase is a pre-processing phase, which includes medical datasets cleaning and continuous feature discretization processes. The next phase is feature reduction of medical data sets using the RS dynamic reduction algorithm. The final phase for classification and medical rules generation for medical treatment decisions. The results showed that classification accuracy for CAD is 91.7% and the accuracy of HCV classification is 85.7%.Next, loosely coupled RS and ANN model is developed. The proposed model used RS in the layers of pre-processing and reduction. Then, it used feed-forward back propagation training algorithm for classification of medical data sets. The proposed model achieved96%and 95% for the classification accuracy of CAD and HCV data sets, respectively.Another model was proposed based on Rough – Genetic algorithm that tries to use the power of RS in data analysis ( preprocessing – reduction – rules extraction) and GA operators as an important tool for rule optimization to maximize the accuracy of the produced decision rules. The results showed enhancement in accuracy for CAD (97.3%) and HSV (96.3%) results.The main proposed model based on Rough Mereology is presented to classify medical data sets that consists of 4 stages in which medical datasets continuous indicators discretized and transformation of attribute data for pre-processing of medical data executed in the first stage, in the second stage Granulation of knowledge using Rough Mereology and Rough inclusion matrix to select the optimum granule radius for rule induction phase that generate a set of probabilistic rules with associated statistical indicators for classification of unseen cases in the final stage. The accuracy of each granule is calculated and used voting by object algorithm to select the best granule. The final results showed that the accuracy of CAD is 96.2 % and96.6% for HSV data set