Abstract
In this study, we present a procedure to help Decision Maker (DM) realize the uncertain situations and provide a quantitative measure to project the DM’s perception to the situations. This procedure takes two types of input: historical data, and the DM’s subjective assessment. Based on this information, possible scenarios can be derived to describe and predict the uncertain situation. When the historical data are not sufficient, in this study, an Analytical Membership Function (AMF) is developed based on a hierarchical clustering procedure; whereas the DM’s perception can be derived from a dialogue with the concept of confidence interval and defined as a DM’s Membership Function (MMF). And then we combine them into a final Integrated Membership Function (IMF) to express the scenarios by an if-then operator. In each membership function, the membership degree expresses the degree of the point in interest belonging to each scenario. The probability of each scenario is determined by a Recency, Frequency, Deviation (RFD) method modified from Recency, Frequency, Monetary method (RFM) for application. In a word, a systematic approach to generate possible scenarios is proposed based on fuzzy set theory when the RFD method is incorporated to derive the probability of each scenario.From a production case of an LED company, we have demonstrated that the generated scenarios can be used in a Robust Optimization (RO) model to help DM yield a robust solution. We have applied the procedure to an LED company and showed the promising results. The result of the illustration showed that this procedure generated three different scenarios of demand with high-, average-, low- demand in IMF. The scenarios could describe all situations of demand via membership function which is not only a better description than the past methodology when facing sparse data; but also give more information. The membership degree represents the degree of demand belonging to a scenario. We defuzzified each scenario from a fuzzy number into a scalar with probability decided by the RFD model. While our procedure predicted the right scenario with 59% of probability with 4.54% in error, the proposed procedure showed the promising results.The proposed procedure can be applied to any enterprise that needs to predict the future situation with little historical data and help DM to make a flexible decision which is precise and with quick response. This study takes a robust optimization problem as an example to illustrate such application possibility.