Development of Decomposition Based Extended Auto-Regressive Integrated Moving Average (DEARIMA) Forecasting Method for Cost Uncertainty

Abstract

Ph.D. Advisor: Professor Kang, Changwook This dissertation addresses the development of a long-term forecasting method that incorporates future uncertainty. To do so, a decomposition time series model has been hybridized with auto-regressive integrated moving average (ARIMA) method to form a new forecasting method. A newly developed forecasting model is called a decomposition based extended auto-regressive integrated moving average (DEARIMA) method. Cost of commodities or processes is a quantitative factor and its fluctuation over the time is hard to predict in a long-term planning horizon. However, DEARIMA has a capability for long-term forecasting under the future uncertainty Cost is a quantitative factor and it can be predicted; however, some other qualitative factors such as response, the obedience of procurement policies, trust, and quality are hard to evaluate or predict. Priority index is introduced in this dissertation for the customer and supplier prioritization based on sustainable qualitative factors. Priority index is a number obtained after the quantification of a set of qualitative factors with some computational procedure. In order to determine the priority index of customers and supplier, an analytical hierarchy process based fuzzy inference decision support system (AHP-FIDSS) is developed. AHP-FIDSS can evaluate the multiple levels of qualitative factors for the inference process. This dissertation is composed of three phases. The first phase is concerned with the development of decomposition based extended autoregressive integrated moving average (DEARIMA) method. In DEARIMA, a novel model selection method called exact combination method (ECM) is introduced. The exact combination method gives the best model parameters for high forecasting accuracy. The introduction of damping factor and fuzziness enabled DEARIMA to be used for the long-term forecasting even under future uncertainty. The second phase of the dissertation is concerned with the development of an analytical hierarchy process based fuzzy inference decision support system (AHP-FIDSS). In AHP-FIDSS, the building blocks are fuzzy inference systems (FIS). In FIS, it is hard to deal the logical rules for more than five qualitative/linguistic variables by experts. However, AHP-FIDSS can be employed for the higher number of variables and their levels. AHP-FIDSS involves the factors screening, hierarchy structure modeling for the quantification of a set of qualitative factors, and their conversion to quantitative values. This phase consists of four steps. The first step of the third phase is concerned with the application and validation of DEARIMA and AHP-FIDSS with the help of case studies. DEARIMA has been applied on time series data of electricity and fuel price in Pakistan. For further validation, Korean government reserves have also been forecasted using DEARIMA. In addition, all these case studies have been solved using ARIMA, winter method, and exponential smoothing. It has been found that DEARIMA has less forecasting error than the other methods. In the second step of the third phase, a case study of customers and suppliers prioritization is presented. Sustainable dimensions have been considered as evaluation criteria. AHP-FIDSS has been used for the determination of the priority index of all supplier and customers. The priority index showed the ranking of suppliers and customers based on their qualitative characteristics related to sustainability. Highly preferred customers or supplier gained high priority index and vice versa. The third step of the third phase includes an example of a multi-period and multi - objective logistic network design with the objectives of cost minimization, greenhouse gas (GHG) emission minimization, and priority index maximization. In the cost function, DEARIMA forecasted values act as a parameter, and, in a priority index function, the parameter values come for AHP-FIDSS. The fourth step of the third phase involves the development of a self-weight adjusting interactive multi-objective fuzzy programming. The proposed method is the modified form of the interactive multi-objective fuzzy programming. Finally, in this phase, a case study of a logistic network is presented for the real-time implementation of the proposed research.