Using neural networks for generalization problems

This paper investigates if neural networks can be used for generalization problems. Generalization problems can be solved by conventional mathematical models or rule-based expert systems if the underlying application domain has complete or partial models. But it is difficult to solve generalization problems, when the problem domain lacks a domain model (we name those as non-conservative domains), e.g., the problem of assigning ratings to corporate bonds. In this paper we explore the application of neural networks in such non-conservative domains. We choose the ratings of corporate bonds as the practical domain for this study because of its enormous importance in the real world of finance.