LSTM Ensemble Approach for Demand Forecasting in Supply Chain Management
Palavras-chave:
Long Short-Term Memory, Ensemble, Autoregressive Integrated Moving Average, Forecasting, Supply Chain ManagementResumo
Perishable product handling represents great challenge to supply-chain management. These items
have shorter shelf-life and most of them have special needs for transportation, storage and display. Inappropriate
treatment of these products, as well as the inability of the retailer to sell them within its shelf-life leads to waste
of products that affects the entire supply chain, society and environment. The trouble to manage these products is
to match the stock replenishment with its future demand and retail profits are significantly affected by perishable
losses. This paper proposes an approach of an one-step ahead forecasting for single time series using LSTM
Ensemble. The forecasts obtained by the proposed method yielded smaller forecasting error compared to the best
single LSTM, best single ARIMA and to the 12-month average sales - method commonly used by retail.
