FUZZY NEURAL NETWORK WITH APPLICATION TO DECISION SUPPORT SYSTEM

Abstract

The main thrust of this thesis is the design of FUZZY NELONET DSS or FUZZY NEural LOgic NETwork based Decision Support System [Nah F.H. (1991)] which incorporates a neural network approach, Fuzzy Neural Logic Network [Hsu L.S., et. al. (1989a, 1989b, 1990a, 1990b, 1990c, 1990d)], into the three components of a decision support system [Sprague R.H.J. (1980)) [Davis M.W. (1988)) [Turban E. (1988)), namely, the data subsystem, the model subsystem and the dialogue or user interface subsystem. Fuzzy Neural Logic Network is both a generalisation and an extension of Neural Three-valued Logic Network [Chan S.C., et. al. (1987, 1988, 1989)). As human reasoning usually involves a certain degree of bias, fuzziness or uncertainty, the three-valued logic concept in Neural Three-valued Logic Network would not be representative of the real world. As such, its representation is extended to fuzzy-valued logic in Fuzzy Neural Logic Network. Two construction algorithms [Nah F.H., et. al. (1990, 1991)], namely, the Local Construction Algorithm and the Global Construction Algorithm, are proposed in this thesis to set up Fuzzy Neural Logic Network by the assignment of connection weights to the neural network. Fuzzy Neural Logic Network can recognise and hence, differentiate all the initial examples even if they are very similar. In addition, conclusions can be made despite minor variations in its inputs. As Fuzzy Neural Logic Network is massively parallel, decisions can be made at high speed. By taking into account the cascade feature of Fuzzy Neural Logic Network, FUZZY NELONET DSS makes decision in a piece meal manner, i.e., once a sub-conclusion is made, other relevant inputs are sought to make decision for the next level and so on. A FUZZY NELONET Developer's Toolkit is developed to serve as a general purpose decision support system shell for FUZZY NELONET DSS. It provides a natural language interface (instead of using conventional binary representation in its interface) with the help of a high-level language known as the Neural Network Design Language [Chan S.C., et. al. (1990)).