Atul Sharma
Production Engineering
July 2025
In the modern competitive production environment, an adequate supply chain resilience and effective operation requires practicable inventory and proper demand forecasting that will allow attaining the desired effect and efficiency. The research uses deep learning (DL) and advanced machine learning (ML) to enhance current planning performance of demand prediction over different planning horizons by using a hybrid ensemble model that combines Long Short-Term Memory (LSTM) networks and Random Forest (RF). The study involved data normalization, dimensionality reduction using Principal Component Analysis model (PCA), and time-series modeling using a real life dataset of 34 months data of 110 product series across online and offline retail channels. The ensemble LSTM + RF model developed outperformed simple methods like ARIMA, ARIMAX, and ML algorithms used individually. The evaluation statistics in terms of Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Error (ME) proved the robustness of the model in terms of the short and medium-term forecasts. The statistically validated predictive accuracy and directional consistency of the ensemble model was also increased by means of statistical tests such as DieboldMariano (DM) and PesaranTimmermann (PT). Also, the model demonstrated stability in the aggregated (monthly) and disaggregated (weekly) forecast and between the retail channels with different volatility of demands. The proposed research develops an intelligent and scalable forecasting framework that facilitates the making of strategic inventory decisions in the production sector to allow improved levels of responsiveness to the customer needs and the operational uncertainties
179- 198
© Airo Journals 2021
