A Study on Demand Forecasting in Supply Chain Management Using Different Deep Learning Approaches

Abstract

In today s competitive market, accurate demand forecasting plays a critical newlinerole in optimizing supply chain management by ensuring product availability while newlineminimizing costs. This research explores the efficacy of various deep learning and newlinehybrid approaches for demand forecasting using publicly available datasets, including newlinethe Corporación Favorita Grocery Sales Forecasting dataset, M5 Forecasting - newlineAccuracy dataset, Rossmann Sales and UCI Online Retail dataset. The study newlineinvestigates the performance of hybrid techniques, combining traditional statistical newlinemodels with advanced machine learning and deep learning approaches. newlineThe Hybrid Prophet- LSTM Approach to improve time series demand newlineforecasting by leveraging the complementary strengths of LSTM (Long Short-Term newlineMemory) networks and Prophet models. LSTM excels at capturing complex, nonlinear newlinetemporal dependencies, while Prophet efficiently identifies seasonal patterns, newlineholidays, and trend changes in the data. The proposed model demonstrates significant newlineimprovements with RMSLE scores across the three datasets, achieving over existing newlinehybrid approaches making it a strong solution for forecasting, particularly with newlineperishable and mixed-type datasets newline