Analisa Dan Implementasi Long Short-Term Memory (LSTM) Dalam Kebutuhan Persediaan Barang di PT. Gunung Sari Indonesia

Abstract

Inventory management plays a vital role in maintaining smooth distribution and operational efficiency within companies. Inaccurate forecasting of inventory needs can cause over-stock or less-stock conditions, leading to increased costs and reduced customer satisfaction. This study applies the Long Short-Term Memory (LSTM) method to forecast inventory requirements based on historical sales data at PT. Gunung Sari Indonesia and compares it with the conventional Moving Average approach. The dataset includes sales transactions from January 2023 to December 2024. The research stages involve data preprocessing, LSTM model construction using window sizes of 14, 30, and 60 days, and performance evaluation using Mean Absolute Deviation (MAD), Mean Squared Error (MSE), and Mean Absolute Percentage Error (MAPE). The results indicate that the LSTM method is more adaptive to fluctuating sales patterns, while the Moving Average method provides more stable predictions for consistent sales patterns. The best MAPE values for the LSTM model range between 102–106%, while the Moving Average method yields values between 85–88%. Therefore, LSTM is preferable for datasets with irregular patterns, whereas Moving Average is more appropriate for stable sales trends.