Fuzzy forecast combiner design for fast fashion demand forecasting

Abstract

In this study, a combiner method is developed to create weekly demand forecasts for a fast-fashion apparel company. The combiner generates forecasts by combining the forecasts of three different methods through fuzzy logic. The combination weights are adaptive in the sense that the weights of the better-performing methods are increased over time. One of the three methods, which is based on product lifecycle, is relatively novel. This method is observed to be quite successful in forecasts as it can reflect the inherent regular seasonality of demand, and it allows the input of expert knowledge. The approach is illustrated through a simulation study that uses real (distorted) data from a Turkish apparel company. The combined forecast method is shown to be better than any of the methods alone.