A new methodology for integrated object-oriented expert systems development in business forecasting

Abstract

The present research is an attempt to tackle some of the existing problems of using expert systems in business forecasting. A case study at the Nippon Electric Corporation (NEC) reveals that the company shares similar business forecasting characteristics as those companies under review. Furthermore the case study indicates that effective business forecasting expert systems require the smooth integration of statistical data and forecasting heuristics capable of delivering accurate and acceptable forecasting results. However, an extensive literature survey done by this author suggests that current business forecasting expert systems do not fully satisfy NEC's requirements. The unsatisfactory state of business forecasting expert systems can be illustrated by the shortcomings of the various approaches. There is a lack of forecasting paradigms that can consistently produce accurate forecasts. The literature survey also reveals that there is insufficient evidence of a formal expert systems development methodology for business forecasting purposes that can address the full cycle of system development and provide a consistent underlying representation of quantitative data and heuristics knowledge in order to facilitate the co-design of an expert system and a conventional system. Various enhancements to business forecasting expert systems have been considered including the integration of statistical techniques, decision support systems, neural networks, and the object-oriented approach. None except the object-oriented approach can satis@ the criteria of consistent representation of quantitative data, heuristics knowledge and full development life cycle coverage. To tackle the shortcomings of existing business forecasting expert systems, a new methodology, Methodology for Integrated Object-Oriented Expert Systems Development (MIOOESD), has been developed in this research. It is developed for business forecasting at NEC. MIOOESD covers the full development cycle systematically in procedures, techniques, and documentation. This artefact is akin to a mature methodology like the Structured Systems Analysis and Design Methodology (SSADM). MIOOESD can further facilitate the co-design of an expert system and a conventional system. Using actual data and forecasting heuristics at NEC, a business forecasting expert system (called "Intelligent Forecaster", IF) has been successfully developed and implemented using MIOOESD. Experience gained from implementing IF suggests that MIOOESD is a useful methodology for developing business forecasting expert systems. This research also leads to a better understanding of the object-oriented expert systems approach in business forecasting. MIOOESD has the potential to be extended to the new generation of computing that involves integration of dissimilar components, e.g. expert system/conventional system coupling in other applications, client/server computing, distributed processing, and so on.