A Design of Analysis Model using Feature Weighting on CBR Method

This paper is a principal idea of case-based reasoning to feature weighting. The feature weighting method called CaDFeW (CAse-based Dynamic FEature Weighting) stores classification performance of randomly generated feature weight vectors. Also it retrieve similar feature weighting success story from the feature weighting case base and then designs a better feature weight vector dynamically for the a new input problem while solving the problem. The CaDFeW is wrapper modelbased feature weighting method that uses classifier error rate as evaluation procedure. To explain the results of applications, this paper is introduced a new definition of input dependency of feature relevance and measured the new concept in the application domains. The empirically measured results showed that relative performance of a local feature weighting method to a global feature weighting method. Key-Words: Machine learning, wrapper method, case-based classifier, feature weighting method.