Automating judgmental decisions using neural networks

Abstract

This paper presents a neural network model that simulates a business loan officer. The network is trained by showing financial ratios, past credit ratings, and loan records of a mixed sample of defaulted and non-defaulted companies. Once it is trained, it recommends to grant or deny a loan. The model uses human judgment of an expert as well as mathematical analysis of financial ratios. It includes into consideration the relative importance of different inputs, and the degree of belief in human judgments. An approach is shown, which allows an "explanation" for the decisions made. The results show that a neural network can be a valuable tool in simulating human decision-making, INTRODUCTION:There has been considerable interest in recent years in computer simulation of human thought processes in problem solving. Expert systems are being developed for various problem domains, including medical and business applications. Tltere are many potential benefits in developing expert systems, which simulate human decision-making. Expert systems save time and money spent on decision-making activities, provide decision skills that are in short supply, train new personnel, and monitor decisions made by humans. The expert systems that simulate human decision-making are better and more consistent in their decisions than the humans. They do not suffer from the problems associated with fatigue, inexperience, and the like. There arc several approaches to develop such systems, including rule-based symbdicprocessing, Bayesian statistics, fu7q logic and neural networks. This paper describes a neural network model that simulates a business loan oficer.A neural network is suitable in situations where natural or plausible reasoning is needed and information is partially incorrect.It operates by analogy or pattern recognition, instead of rule-based logical reasoning. It is kest suited for problems where rule sets are unclear, too dynamic, or too large. The model presented here uses backpmpagation neural networks. It has two modules. Each module has a Permission to copy without fee all or part of this material is granted provided that the copies are not made or distributed for direct commercial advantage, the ACM copyright notice and the title of the publication and its date appear, and notice is given that copying is by permission of the Association for Computing Machinery. To copy otherwise, or to republish, requiree a fee end/or specific permission. a 1992 ACM 089791 -472-4/92/0002/0351 $1.50