Optimization of Demand Forecasting in the Supply Chain Management of Apparel Industry

Abstract

Accurate demand forecasting is a key component of a well-built supply chain management process in the ever-changing apparel industry, where precise predictions are vital for optimizing the production, inventory, and transportation levels. Traditional methods on numerous occasions fail to comprehensively understand the nature of this field, thus resulting in inefficiencies within the Sri Lankan apparel supply chain. The research answers this problem through the identification and development of the context-specific methods that are effective for enhanced demand forecasting in the apparel supply chain. The project explores the use of deep learning, particularly Long Short-Term Memory (LSTM) networks and their combinations with other models (CNN, ARIMA, BPNN) to develop a demand forecasting application. Experiments with six models identified a CNN-LSTM architecture as the optimal solution, achieving the lowest MAE of 2.9710, MAPE of 24.6802, MSE of 85.0358, and RMSE of 9.2215. Hyperparameter tuning and cross-validation were employed to optimize and validate the chosen model.