Experts typically have unequal authority levels in collaborative forecasting tasks. Most current fuzzy collaborative forecasting methods address this problem by applying a (fuzzy) weighted average to aggregate experts’ fuzzy forecasts. However, the aggregation result may be unreasonable, hence fuzzy weighted intersection operators have been proposed for fuzzy collaborative forecasting. This paper proposes that unequal expert authority levels are considered when deriving the membership function rather than the aggregation value. Therefore, the membership of a value in the aggregation result cannot exceed those in experts’ fuzzy forecasts. The proposed approach was applied to forecast the yield of a dynamic random access memory product to validate its effectiveness. Experimental results showed that the proposed methodology outperformed all current best-practice methods considered in every aspect, and in particular achieving 65% mean root mean square error reduction. Thus, a high expert authority level increased the likelihood for the forecast, which could not be satisfactorily addressed by simply applying a higher weight to the forecast.
- Dynamic random access memory
- Fuzzy collaborative forecasting
- Fuzzy weighted intersection