Thls paper deals with the use of neural network rule extraction techniques based on the Genetic PFogramming (GP) to build intelli.ffent alld explanatory evaluation systems. Recent development in algorithms that extract rules from trained neural networks enable us to generate classificatlon rules in spite of their intrinsically black-bex nature. However, in the original decompositional method looking at the internal strueture of the networks, the comprehensive methods combining the output to the inputs using parameters are complicated. Then, in our paper, we utilized the GP to automatize the rule extraction process in the trained neural netwerks where the statements changed into a binary classification. Even though the production (classification) rule generation based on the GP alone are applicable straightforward to the underlying problens fbr decisie] making, but in the original GP rnethod production rules include rnany statements clescribed by arithmetic expressions as well as basic logical expressions, and it makes the rule generation process very complicated. Therefore, we utilize the neural netwerk and binary classifieation te obtain simple and relevant classification rules in real applications by avoiding straightforwarcl applieations of the GP procedure to the arithmetic expresslons. At first, the pruning process of' weight among neurons is appried to obtain simple but substantlaL binary expressions which are used as statements is classification rules. Then, the GP is applied to generate uleimate rules. As applications, we generate rules to predictton of bankruptcy and creditwotthiness for binary classifications, and the apply the methocl to multi-level classification of corporate bonds (rating) by using the financial indicators.
SUMMARYThis paper deals with the realization of feature description systems for clusters by rule generation based on genetic programming (GP) and its applications. First, the data are divided into several clusters by using conventional clustering algorithms. Then logical variables corresponding to the categorical variables are introduced, and the logical expressions using these logical variables are defined as rules to extract the targeted cluster from the dataset. The rules are improved by GP so that they are valid (become true) only for the targeted cluster. Unlike ordinary GP procedures, the fitness of individuals is defined as proportional to the number of hits inside the targeted cluster, but also to the inverse of the number of hits outside the targeted cluster. In simulation studies, the system is applied first to artificially generated samples and clusters to examine the performance of the system, and then to personal loan assessment problems, after which the evaluation of several kinds of clustering problems is summarized.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.