Inventory model with demand forecasting based on long short-term memory

This study presents an inventory model that integrates machine learning-based demand forecasting to optimize inventory decisions under uncertain demand conditions. Weekly demand data from weeks 1 to 52 in 2021 were forecasted using Long Short-Term Memory (LSTM) and Bayesian Regression models in Python, as well as Exponential Smoothing (ETS) implemented in Excel. These forecasts were then applied to an inventory simulation using Excel Solver. Among the three forecasting methods, LSTM demonstrated the highest accuracy and was selected for further integration. The simulation compared three inventory control methods combined with LSTM, revealing that Method 3 consistently achieves a 100% Fill Rate and reduces total inventory costs by 13.72% compared to the baseline. While its total cost is slightly higher than Method 2, it remains significantly lower than Method 1, offering a reliable balance between cost efficiency and service level. The findings emphasize the benefits of combining advanced forecasting techniques with appropriate inventory policies to effectively manage demand variability, minimize stockouts, and improve overall supply chain performance.