| Building a good model based on observed data and understanding the patterns generated by the model are two keys issues in data mining. The ability of neural networks to learn patterns from noisy data has allowed them to become a popular tool for data mining.; This study introduces a neuro-fuzzy-genetic data mining architecture, which discovers patterns and represents them in understandable forms. It is an attempt to combine computational intelligence tools: neural networks, fuzzy logic, and genetic algorithms to data mining problems. In the architecture, Principal Component Analysis (PCA) is applied to reduce the dimensions of the input variables in finding combinations of variables, or factors, that describe major trends in the data. The reduced dimensions of input variables are then used to train Probabilistic Neural Network (PNN) to classify the dataset according to the classes considered. A rule extraction technique is then applied in order to extract explicit knowledge from the trained neural networks and represent it in the form of crisp and fuzzy If-Then rules. In the final stage, a genetic algorithm is used as a rule-pruning module to eliminate those weak rules that are still in the rule bases.; Comparison of the architecture with the standard C4.5 decision tree was carried out. The rule bases extracted from the architecture outperform those generated from C4.5 in classification accuracy and the number of extracted rules. The capability of the architecture is demonstrated with four real world datasets from different fields: telecommunications, financial, medical, and marketing. Further considerations of other rule extraction techniques and different neural network structures should be encouraged.*; *This dissertation includes a CD that is compound (contains both a paper copy and a CD as part of the dissertation). The CD requires the following applications: Microsoft Office; Matlab; Winzip; Internet Browser. |
