An Ensemble Approach Combining LSTM and Random Forest for Forecasting Product Stock Expenditure in Retail

Demand forecasting is essential for optimizing inventory management in retail, ensuring product availability while minimizing the risks of overstocking and stockouts. This study explores the application of deep learning and machine learning techniques to improve demand forecasting accuracy, focusing on the integration of Long Short-Term Memory (LSTM) networks and Random Forest models. Following the CRISP-DM (Cross-Industry Standard Process for Data Mining) methodology, the research adopts a structured approach from business understanding to model development. Ensemble methods, incorporating Stacking, Voting, and Bagging, are employed to leverage the strengths of both LSTM and Random Forest models. The results demonstrate that the Random Forest achieving the lowest MSE (74.6), MAE (6.98), and RMSE (8.63). The Bagging model excels in reducing percentage-based errors, yielding the lowest MAPE (80.99%) and SMAPE (20.58%). In contrast, the LSTM model exhibits the weakest performance across all metrics, particularly with a MAPE of 155.1%. Following the modeling phase, the Bagging model was deployed using Streamlit, enabling real-time access to predictions. These findings highlight the effectiveness of ensemble methods, particularly Bagging, in enhancing demand forecasting accuracy. Furthermore, the study underscores the potential for future research to explore hybrid approaches and incorporate additional factors to refine predictive capabilities in retail inventory management.