This paper describes the use of a neuro-fuzzy-genetic data miningarchitecture for finding hidden knowledge and modeling the data of the1997 donation campaign of an American charitable organization. This datawas used during the 1998 KDD Cup competition. In the architecture, allinput variables are first preprocessed and all continuous variables arefuzzified. Principal component analysis (PCA) is then applied to reducethe dimensions of the input variables in finding combinations ofvariables, or factors, that describe major trends in the data. Thereduced dimensions of the input variables are then used to trainprobabilistic neural networks (PNN) to classify the dataset according tothe groups considered. A rule extraction technique is then applied inorder to extract hidden knowledge from the trained neural networks andrepresent the knowledge in the form of crisp and fuzzy if-then-rules. Inthe final stage a genetic algorithm is used as a rule-pruning module toeliminate weak rules that are still in the rule base while insuring thatthe classification accuracy of the rule base is improved or not changed.The pruned rule base helps the charitable organization to maximize thedonation and to understand the characteristics of the respondents of thedirect mail fund raising campaign
展开▼
