Goonatilake, Suran; (1994) An Intelligent Hybrid System for Financial Decision Support. Doctoral thesis (Ph.D), UCL (University College London).

Text An intelligent hybrid system for financial decision support.pdf Download (74MB)

Abstract

Recently there has been much research in using intelligent techniques to assist decision making. It is noted, however, that most intelligent techniques have limitations, and that they are not universally applicable to all decision making tasks. Each intelligent technique has particular computational properties making them suitable for certain tasks over others. Against this backdrop, this thesis puts forward a novel intelligent hybrid systems approach for supporting financial decision making. An important contribution of the thesis is the formulation of a new classification scheme for intelligent hybrid systems, which takes into account factors such as functionality, processing architecture and communication requirements. The three proposed classes of hybrid systems are Function-Replacing, Intercommunicating and Polymorphic. This classification provides a mechanism to make qualitative assessments of existing hybrid systems and also helps to guide the development of new hybrid architectures. The key requirements of a financial decision support system have been analysed in-depth. It is argued that an ideal decision support system for financial decision making should satisfy a range of criteria including: the ability to induce decision making knowledge from domain data; the ability to process fuzzy relationships; the ability to adapt to changes in the decision making environment; the ability to provide explanations and the ability to allow decision makers to change and add new knowledge. A hybrid system which demonstrates these computational concepts, INTENT, is developed and its effectiveness demonstrated in the complex decision making task of foreign exchange trading. INTENT combines expert systems, fuzzy systems, genetic algorithms and neural networks to produce an effective decision support system. All decision models derived from the system have been critically evaluated by a domain expert. This provides a mechanism to assess the effectiveness of the approach in terms of its explanation capabilities and the ease of judgmental revisions to decision models. The quantitative performance of the approach has been assessed against the decision-making performance of a human trader, and has also been compared with results in the trading systems literature. With respect to data pre-processing, a novel clustering-based method for converting raw domain data into symbolic linguistic descriptions is developed. This method is further extended to convert raw domain data into fuzzy logic descriptions by finding appropriate fuzzy membership functions. A method of using genetic algorithms to induce fuzzy models is also developed and its effectiveness is demonstrated. Based on the experiences obtained during this project, a preliminary scheme for hybrid systems development is proposed. This scheme draws upon our three hybrid classes and offers a set of guidelines based on the problem domain and assessments of computational properties of different intelligent techniques.

Download activity - last month

Export as JSONCSVXML

Download activity - last 12 months

Export as XMLCSVJSON

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item